The White Working Class and Voter Turnout in U . S . Presidential Elections , 2004 to 2016

Through an analysis of the 2004, 2008, 2012, and 2016 Current Population Surveys as well as the 2004 through 2016 General Social Surveys, this article investigates class differences and patterns of voter turnout for the last four U.S. presidential elections. After developing some support for the claim that a surge of white, working-class voters emerged in competitive states in 2016, a portrait of class differences on political matters among white, non-Hispanic, eligible voters between 2004 and 2016 is offered to assess the electoral consequences of this surge. These latter results are consistent with the claim that racial prejudice, anti-immigrant sentiment, concerns about economic security, and frustration with government responsiveness may have led many white, working-class voters to support an outsider candidate who campaigned on these themes. However, these same results give no support to the related claim that the white working class changed its positions on these matters in response to the 2016 primary election campaign or in the months just before the general election.

I N the hours following the 2016 U.S. presidential election, pundits and pollsters constructed a white working-class narrative to explain why Donald Trump defeated Hillary Clinton.The New York Times, in its lead story the morning after the election, wrote that Trump's victory "was a decisive demonstration of power by a largely overlooked coalition of mostly blue-collar white and working-class voters who felt that the promise of the United States had slipped their grasp amid decades of globalization and multiculturalism" (Flegenheimer and Barbaro 2016).Front-page headlines included "Working Class Speaks" and "Blue-Collar Whites Give Stinging Rebuke to Democratic Party."The online version of the same article linked directly to a piece by the newspaper's most prominent data journalist, Nate Cohn, entitled, "Why Trump Won: Working-Class Whites" (Cohn 2016).
The white working-class narrative, if true, is said to account for the unexpected breach of the Democratic "blue wall" states of Pennsylvania, Michigan, and Wisconsin as well as Trump's more decisive victories in Florida and Ohio.It does have some face validity.Donald Trump had indeed called for a revitalization of working-class economic security through renegotiated trade agreements, reductions in immigration from Mexico, status-quo funding for Social Security, and (during the campaign) a middle-of-the-road position on health care reform.He also delivered this bundle of policy priorities in effervescent rallies in competitive states that he then carried in the primary and general elections.Embracing his totemic red hat, these rallies provided suggestive evidence for a secondary and more controversial component of the white working-class narrative: that anti-immigrant rhetoric and "dog whistle" warnings of racial threat can motivate white voters to support candidates who oppose increases in multiculturalism and that a disproportionate share of these sufficiently impressionable white voters are located within the working class.
The initial case for the white working-class narrative was based on one uncontestable fact and one clear pattern that emerged from exit poll data on Election Day.For the first, Trump won four rustbelt northern states-Pennsylvania, Ohio, Michigan, and Wisconsin-that Barack Obama had carried in both 2008 and 2012.For the second, more white voters without a college degree claimed to have voted for Trump in 2016 than was the case for Romney in 2012.On the basis of their analysis of exit poll data, the team of data journalists at the New York Times reported that 14 percent of white voters without a college degree had switched from voting Democrat to voting Republican between 2012 and 2016 (see Huang et al. 2016). 1 In this article, we seek to contribute to the ongoing evaluation of the white working-class narrative by using measures and data sources that were unavailable to pundits and pollsters in the weeks and months just after the election.Our most important contribution is an analysis of changes in voter turnout rates using a genuine measure of social class with the Current Population Survey Voting and Registration Supplements (CPS-VRS).Because the CPS-VRS does not include any measures of political or social attitudes, we also draw on the rich information collected for the General Social Surveys (GSS) in order to consider the portions of the explanation that are based on conjectures about class-specific economic interests, anti-immigrant sentiment, racial prejudice, and frustration with the responsiveness of government.Before offering our analysis, we summarize what has been learned about actual voting patterns since the white working-class narrative was first constructed in November 2016.

Vote Tallies and Geographic Variation
The sanctity of the voting booth prevents any direct analysis of how cast votes are related to the characteristics of individual voters, but other types of indirect analysis are possible.Precinct totals can be tabulated from official results and aggregated to larger geographic units.Then, variation in these totals can be related to aggregate measures of individual characteristics that are calculated with data from the U.S. Census Bureau.This sort of analysis, conducted by data journalists in the weeks following the election, revealed patterns that were mostly consistent with the white working-class narrative.Counties in which Trump gained votes in 2016 relative to Romney in 2012 tended to have larger relative shares of white residents with low average levels of education (see Silver 2016).
Although these findings are supportive of the initial interpretations of exit poll data, it is important to appreciate their limitations. 2Level of education completed is not a direct measure of class, and the larger vote shares captured in flipped "Trump counties" were more prevalent in rural areas.Most white, working-class voters do not live in rural counties, and aggregate county-level analyses can yield misleading conclusions about individual-level patterns and generate what social scientists have labeled ecological fallacies of inference.

Votes Cast, Based on Retrospective Self-Reports
The traditional method that university-based researchers use when analyzing votes is to ask samples of eligible voters whether they voted and, if so, for whom they voted.For decades, researchers have relied on the American National Election Studies (ANES), which survey a national sample of U.S. citizens before and after each presidential election using in-person interviews.For the 2016 election, two surveys of more recent vintage have also received a good deal of attention, especially among journalists: (1) a poll of adult Internet users undertaken by YouGov for the Democracy Fund Voter Study Group and (2) a telephone-recruited but web-based monthly panel survey, the American Trends Panel, conducted by the Pew Research Center. 3uch research is ongoing with these data sources, and all of this research points to a common third finding that is also consistent with the white working-class narrative: a meaningful proportion of self-identified Trump voters reported that they had voted for Obama in a prior election.Evidence is mounting that these Obama-to-Trump voters are disproportionately white and have lower levels of education.Our expectation is that this finding is likely to withstand inevitable attempts to debunk it in the coming months and years.
Nonetheless, it should be recognized that this finding also provides no direct support for the white working-class narrative.None of these surveys or polls collects information on occupation, and thus, none enable a direct analysis of shifts in the support of the working class relative to other shifts in support. 4These surveys also show every other conceivable pattern of switching, such as nontrivial shares of Romney-to-Clinton voters and so forth.We have yet to see an analysis that considers all shift patterns between 2012 and 2016 decomposed with a direct measure of class.
Voter Turnout, Also Based on Retrospective Self-Reports We know from tabulations of actual votes counted as well as reliable estimates of the population of eligible voters that the turnout rate for the 2016 presidential election was about 60 percent of eligible voters. 5Unfortunately, none of the data sources just discussed can be used to estimate voter turnout to an acceptably accurate degree.Exit polls include information only on those who voted, and the other surveys yield self-reported turnout rates that are implausibly high.The most extreme is the Democracy Fund Voter Study Group's YouGov poll of Internet-using adults, which, for example, yielded a turnout rate for 2016 of 92.7 percent.In addition, the calculable turnout rate for the YouGov poll differed little by the respondents' selfreported level of education-from a high of 97.6 percent of those with a graduate degree to a low of 88.7 percent of those with a high school diploma or less (see  Democracy Fund Voter Study Group [2017]: Table 2).These are implausible turnout rates that demonstrate why this poll cannot be used to estimate actual turnout.
The usual interpretation of upward biases like these is twofold. 6First, individuals inclined to vote are more likely to agree to participate in surveys and polls that they are informed concern political matters.Second, for panel surveys with preelection data collection, participation itself heightens interest in political matters. 7or this second reason, it is thought that a substantial number of respondents who decide to vote would not have voted if they had not been a participant in the study.For these reasons, in order to study voter turnout, it is preferable to use a more general survey that mitigates these response dynamics, and the favored choice is the CPS-VRS, which we analyze in this article. 8

Plan of Analysis
Analyzing the 2004, 2008, 2012, and 2016 CPS-VRS, we will first consider the relationship between class position and turnout rates in each of the last four presidential elections.The Voting and Registration Supplement to the regular, monthly CPS is conducted in the weeks immediately after each year's November election. 9It is the only data source with information on voter turnout that (1) has occupation measures that enable class coding and (2) has a sufficient sample size to reliably disaggregate turnout by class and geographic region.After analyzing the CPS-VRS data to consider whether turnout rates increased in 2016 among white, workingclass voters in competitive states, we then turn to an analysis of the 2004 through 2016 GSS in order to investigate the social and political attitudes of the white working class in comparison to white-collar workers and others. 10We conclude with a consideration of how the overall composition of voters changed between 2012 and 2016 in competitive states and show how baseline changes in the distributions of race-ethnicity and class membership combine with turnout rates to determine the composition of those who cast votes.

Data and Measures
For both the CPS-VRS and the GSS, we selected subsamples of eligible voters only (including individuals who are eligible to vote but are not registered). 11In the online supplement, we provide details on the construction of the CPS-VRS analysis sample as well as an explanation of our implementation of a weighting procedure proposed by Hur and Achen (2013) to better align the CPS-VRS with known vote totals across states.For the GSS analysis, we enacted the same basic sample-construction decisions chosen for Morgan and Lee (2017).Additional details for the GSS sample are also provided in the online supplement and, more generally, in Marsden (2012) and Smith et al. (2017).

Coding of Race and Ethnic Self-Identification
In the years following the 2000 U.S. Census, both the CPS and GSS have allowed respondents to select multiple categories when expressing their racial and ethnic selfidentification.The particular categories offered continue to evolve, and collectively, they are still too constrained to represent all patterns of interest to all respondents.Nonetheless, both surveys elicit responses that allow for consistent measurement of the "white" portion of the white working class.
For this article, we define the particular whiteness of concern in the white working class to be "white only and non-Hispanic."Respondents in this category indicate that they do not consider themselves to have any type of Hispanic, Latin American, or Chicano ancestry when responding to the Hispanic ethnicity question, and they select only "white" from among the options for race.Accordingly, multiracial whites and white Hispanics are both excluded from the constructed category of "white" in this article.
We recognize that this analysis decision will be objectionable to some readers.Our position is that this measurement decision is most consistent with the white working-class narrative that we aim to evaluate as well as with the other polls and surveys that are being analyzed by other researchers concurrently.For other research projects that utilize the CPS-VRS or the GSS, alternative operational definitions of "white" would be more appropriate.
In addition, we include all other eligible voters in an omnibus "all other" category for race-ethnicity.The focus of this article is the white working class for the reasons already stated, and a full analysis of variation attributable to all of the ethnoracial diversity present in the CPS-VRS and GSS is beyond the scope of this article.In addition, we want to avoid focusing on only the one or two additional groups that can be more easily measured because of their comparatively large size: eligible voters who are black or African American and eligible voters who are non-black and Hispanic.This decision allows us to avoid contributing to the marginalization of smaller groups.Our choice is to retain all eligible voters in our analysis and focus our interpretations on the group most central to evaluating the white working-class narrative.

Coding of Class
We adopt a coding for class based on the 2000 and 2010 U.S. Census occupational classifications.Descriptions of the classes are presented in Table 1, and the coding is based on the employment relations perspective that is elaborated in cross-national work on social stratification and class voting (see Erikson and Goldthorpe 1992;  Evans 1999).For more details on the class coding for the CPS-VRS, see the online supplement and Morgan (2017). 12able 1 presents the underlying classes that the coding yields, categorized into four groups that will represent our most prominent categorization of eligible voters in this article (but see the online supplement for parallel results that disaggregate all classes when sample size constraints allow).We consider classes I, II, and IIIa to be a class group that is not working class.This group includes professionals and other highly skilled workers in classes I and II as well as office-based clerical and health workers, who usually have some higher education, in class IIIa.For many projects, variation across these three classes is crucial to consider.For this article, this group will serve as our primary reference group for comparisons with the working class.
Our working-class group includes lower-grade service workers (class IIIb) as well as manual workers, both skilled (class VI) and unskilled (class VIIa).We also consider two other class groups, neither of which is clearly working class or not.The first is an intermediate class group that includes self-employed nonprofessional workers (class IVab) as well as higher-skilled manual workers and supervisors (class V), the latter of which includes public safety workers and others whose conditions of Notes: For all detailed occupations assigned to each class, see the online supplement.For the rationale for the assignments, see Morgan (2017).
employment are typically more favorable than those in working classes IIIb, VI, and VIIa.The final group is composed of farmers, ranchers, and farm managers (class IVc) as well as agricultural workers (class VIIb).These last two underlying classes differ from each other in many ways.But, when restricted to non-Hispanic whites, their political behavior and attitudes are more similar than one might otherwise expect, perhaps owing to their common place of residence.

Coding of Competitive and Noncompetitive States
Turnout is known to be lower in noncompetitive states during presidential elections, and most explanations attribute this difference to voters' perceptions that their votes are less important in determining the outcome.Presidential candidates also do not campaign in the same way in noncompetitive states, and they commit fewer resources to their turnout ground game.For this reason, for our CPS-VRS analysis we will consider only competitive states for our main results (but see the online supplement for parallel results for all states).These 18 states are as follows: Arizona, Colorado, Florida, Georgia, Iowa, Maine, Michigan, Minnesota, Missouri, Nevada, New Hampshire, New Mexico, North Carolina, Ohio, Pennsylvania, Texas, Virginia, and Wisconsin.
Based on publicly available vote totals, these are the states where the margin of victory was 10 percent or less in either the 2012 presidential election or the 2016 presidential election.These states include bellwether battleground states, such as Florida and Ohio, but also the Democratic blue wall states of Pennsylvania, Michigan, and Wisconsin.And they include states with more rapidly changing demographic profiles, such as Arizona, Colorado, Nevada, and Texas.Our analysis of the CPS-VRS indicates that 47.4 percent of eligible voters resided in these competitive states in the 2016 election.(Some of these 18 states were not competitive in prior elections, such as Georgia and Texas.Partly for this reason, our turnout rate results for 2004 and 2008 are interpreted mostly for context given that our definition of competitiveness was chosen to facilitate a comparison of 2012 and 2016.) For our subsequent GSS results, we analyze a national sample.The GSS is not representative at the state level, and the sample size of the GSS is far smaller.In addition, there is no reason to expect that the attitudes we model for that portion of the analysis are particularly sensitive in any direct way to the ground game of alternative political parties, unlike turnout, which is known to vary by the competitiveness of the state, especially from 2004 onward (see Green and Gerber  2015).

Results
We offer results in three separate sections.For the first, we present estimates of voter turnout rates by class for the last four presidential elections within 18 states that were competitive in 2012 and 2016.Our goal is to assess whether turnout rates increased for the white working class in 2016 in comparison with 2012.We also provide estimates for 2004 and 2008 in order to provide context.In the second section of our analysis, we then develop a portrait of class differences in attitudes that are central to political matters thought to be relevant to the white working-class narrative and how these attitudes changed between 2004 and 2016.Finally, in the third section, we return to the results on voter turnout and consider compositional changes among voters from 2012 to 2016.

Turnout Rates by Class and Race-Ethnicity
For voter turnout research, the pool of eligible voters includes those who voted, those who were registered to vote but did not vote, and those who were eligible to vote but were not registered.Before presenting our analysis of turnout patterns, we consider basic characteristics of the pool of eligible voters in 2016.
For the 2016 election, the CPS-VRS indicates that 69.3 percent of eligible voters self-identified as non-Hispanic and white only.This percentage has fallen steadily across the last four presidential elections from 75.5 percent in 2004.In the competitive states that we have selected for our primary analysis, the corresponding percentages are slightly higher at 70.8 percent in 2016 and fell gradually from 77.7 percent in 2004. 13able 2 presents the joint distribution of race-ethnicity and class in 2016 for eligible voters living in the 18 competitive states.For CPS-VRS sample members who were out of the labor force, no information on occupation was collected.It is impossible to assign a class to these sample members even though many of them still identify with an occupation and thus would be able to report a most recent occupation if asked. 14To create an exhaustive classification of all eligible voters sampled for the CPS-VRS, we therefore allocated individuals without an assigned class to three additional groups: those with a bachelor's degree or higher, those with more than a high school diploma but without a bachelor's degree, and those with a high school diploma or less. 15he marginal distribution of class and education groups in competitive states for 2016 is presented in the final column of the first panel of Table 2.As shown there, 20.5 percent of eligible voters in competitive states were currently employed in the working-class group (classes IIIb, VI, and VIIa), with another 8.7 percent employed in the intermediate-class group (classes IVab and V).Only 0.7 percent were employed in farming or as agricultural workers (classes IVc and VIIb).These three groups together are slightly larger than the percentage of employed whitecollar workers, which is 27.0 percent.Finally, the largest group of eligible voters without a class position is the 20.8 percent of eligible voters who have no more than a high school diploma.An additional 13.4 percent had at least some postsecondary education.Many of the eligible voters in these two groups would be members of the working class if they had been asked about their most recent occupation.
Other percentage comparisons within the columns and rows of Table 2 are consistent with the structure of inequality and patterns of enfranchisement.Among non-Hispanic whites, 18.6 percent were employed in the working class in comparison with 25.3 percent of all others.Still, because of relative size, non-Hispanic whites represent 64 percent of the employed working class.In fact, because of their prevalence among eligible voters, non-Hispanic whites are a clear majority of all class and education groups in Table 2. Notes: CPS-VRS sample members, weighted by using the CPS-VRS variable "pwsswgt" (including sample members who are eligible to vote but are missing information on the voting questions).The raw N for the table is 34,659 (the weighted N for comparison with the other tables is 43,968).
Figure 1 presents our key results for turnout in competitive states, in which the dots are point estimates of turnout rates and the vertical bars are plus-or-minus one standard error for each estimated rate.We utilize two separate panels for clarity, and they are scaled in the same way.As shown in the first panel, the white-collar group has the highest turnout rate, oscillating between 80 and 82 percent for non-Hispanic whites (in red).For all others (in gray), the turnout rate has an inverted U-shaped pattern, with a peak in 2008 of 76.6 percent and a decline through 2016 to 70.2 percent.
As shown in the same panel, the turnout rate for the working class was considerably lower.For non-Hispanic whites in the working classes of IIIb, VI, and VIIa, the turnout rate increased from 53.5 percent to 56.9 percent between 2012 and 2016.Coupled with the decline for all others from 53.4 to 49.5 percent between 2012 and 2016, the race-ethnicity gap in the turnout rate within the working class was  larger in 2016, at 7.3 percent, than in the three prior elections.It is notable that the next-highest gap among the other three elections was observed in 2004, which was the last victory by a Republican presidential candidate.For 2004, the gap was 5.5 percent (i.e., 55.8 percent compared with 50.3 percent) in these same states.
The second panel gives corresponding turnout rates for the other two class groups (although we suppress the "all other" farmer and agricultural worker turnout rate because the sample of eligible voters is too small to yield meaningfully precise estimates for this group). 16Here, the intermediate-class group shows a widening within-class race-ethnicity gap, but overall, only a very small turnout rate increase for non-Hispanic whites in 2016 is evident.And, consistent with journalistic analyses of the rural counties discussed in the introduction, the turnout rate for non-Hispanic white farmers and agricultural workers in classes IVc and VIIb increased from 64.1 percent to 74.2 percent from 2012 to 2016 (albeit with much larger standard errors of 3.7 and 3.6 percent, respectively, because of the sample size).
To complete our analysis of turnout rate patterns in competitive states, Figure 2 presents the rates for the education groups introduced above in Table 2.These are large groups because they include all eligible voters who were out of the labor force or retired.For these groups, turnout rates evolved in patterns somewhat consistent with the class differences presented in Figure 1.In particular, turnout rates increased  to at least some extent for all groups of non-Hispanic whites between 2012 and 2016 on an absolute scale and especially in relative comparisons to all others in the same education group.The largest of these groups, which is composed of those out of the labor force and with no more than a high school diploma, shows a trivially small increase in the turnout rate among non-Hispanic whites.However, a turnout rate gap nonetheless emerged within this group for 2016 because of a substantial decline in the turnout rate for all others.
Altogether, the CPS-VRS offers some support for the white working-class narrative.Turnout rates increased between 2012 and 2016 by 3.4 percent for non-Hispanic whites in the working class, and the impact of this increase was enhanced by a simultaneous turnout rate decrease of 3.9 percent among all other eligible voters in the working class.In addition, the CPS-VRS suggests that the turnout rate increased even more among non-Hispanic whites who worked as farmers, ranchers, or other agricultural workers, and this pattern is consistent with Trump's notable strength in rural counties in competitive states.
We will discuss these turnout rate patterns in more detail below when we consider overall compositional changes between 2012 and 2016 and when subsequently synthesizing our results.Next, we turn to an analysis of the GSS and seek to determine whether these turnout rate changes are consistent with the explanatory core of the white working-class narrative.We examine whether class differences are present in relevant political attitudes and whether members of the white working class, on average, changed their positions in 2016.

Class Differences and Changes in Attitudes as Measured by the General Social Surveys
We first examine opinions on government responsiveness, self-avowed understanding of political issues, and the role of the government in the economy and the provisioning of the social safety net.We focus on class differences among non-Hispanic white, eligible voters only.The goal is to better understand whether the turnout rate changes presented above are consistent with the white working-class narrative.
The analysis in this section is based on a comparison of two comparable government responsibility modules fielded for the GSS in both 2006 and 2016. 17Before presenting the results, we should establish two points about the over-time comparison and the class coding for the GSS.
First, 2006 and 2016 constitute a reasonable comparison for an assessment of change in these items, which include opinions on the stewardship of the economy and the need for the social safety net.In both years, during the months when the GSS was fielded, the unemployment rate was low by historical standards even if the Great Recession was still fresh in the minds of many 2016 respondents. 18In other words, the consequences of a key determinant of current feelings about the role of government-the current state of the economy-is mitigated by a comparison of these two years.
Second, unlike the CPS-VRS, the GSS elicits each respondent's current or most recent occupation, which means that the GSS collects information on the class position of those who are not currently in the labor force, which is most commonly the case because the respondent is retired.We see this as a key advantage of the GSS insofar as we adopt the position that the most recent occupation held is likely still a source of identification among those not currently employed.We assume, for example, that retired lawyers and retired carpenters can be reasonably categorized as members of classes I and VI, respectively. 19esponsiveness of the government and understanding of political matters.In both 2006 and 2016, GSS respondents were presented with two items: Please tell me how much you agree or disagree with each of the following statements: People like me don't have any say about what the government does.
I feel that I have a pretty good understanding of the important political issues facing our country.The numbers of respondents are 1,673 (left) and 1,671 (right), with the variation in sample size reflecting different rates of "don't know" and refusals for the two items.
two instances).The estimated percentage rates are adjusted by respondent age, following the public opinion tradition that stresses the utility of interpreting crosssectional differences only after adjusting for differences between age cohorts. 20ith the modest sample sizes available for the GSS (about 850 non-Hispanic white, eligible voters per year for the items we consider in this section), this adjustment is also helpful in stabilizing between-class and between-year comparisons.The results in Figure 3 reveal class differences in feelings toward the government but little evidence of change.(In addition, see the disaggregated results for each underlying class in the online supplement.The results presented there reveal a wider range of class differences, such as differences between class II and class VI, but also offer no additional evidence of change.)The first panel of Figure 3 shows that the working class feels comparatively powerless in political matters, with more than half of all respondents reporting that people like them "don't have any say about what the government does."In comparison, the white-collar group has somewhat greater confidence in its political influence, and the intermediate-class group is midway between the white-collar group and the working class.Finally, the comparatively small group of farmers and agricultural workers appears to feel the most powerless, but estimation error is regrettably large for this group, preventing any clear conclusion.
The second panel of Figure 3 shows a closely related pattern, in which those class groups that feel the most powerless are also the groups that feel that they have the least-developed understanding of the political issues facing the country.However, confidence is nonetheless high, with all groups well above 50 percent in expressing confidence that they understand the most important political issues facing the country.
Before carrying on to additional attitude items, we should concede that there are two ways to interpret the over-time differences in Figure 3.A cautious interpretation would suggest that we cannot say much of anything about change over time given the width of the error bars associated with each point estimate.The less cautious interpretation is the one that we favor: even with independent samples of less than 1,000 non-Hispanic white GSS respondents per year and with each sample selected a decade apart from the other, the point estimates for each class group line up in rather close proximity for both 2006 and 2016.So although it is possible that much larger samples than the GSS could detect genuine changes between 2006 and 2016, we see the GSS as providing meaningful evidence that, on these two items, little has changed in the pattern of class differences.
Material interests and the role of the government in the economy.In Figure 4, we examine four items on government responsibility and management of the economy.GSS respondents were asked: On the whole, do you think it should or should not be the government's responsibility to . . .Reduce income differences between the rich and the poor?Keep prices under control?Provide a job for everyone who wants one?Provide industry with the help it needs to grow?
Respondents were provided with four response options: "definitely should be," "probably should be," "probably should not be," and "definitely should not be."The percentages in Figure 4 are for those who selected "probably should be" or "definitely should be."We again adjusted for age cohort, as done for the results presented in Figure 3.
Across all four items, working-class respondents favored more activist government intervention.They are more likely to see the government as responsible for reducing inequality between the rich and poor and providing jobs for those who want them.They also favor effort to control inflation, and they support providing help for industry to grow.Like all other respondents, working-class respondents were more likely to favor government effort to control inflation and support industry than they were to favor government effort to reduce inequality and provide jobs.In other words, although working-class respondents would appear to favor their own material interests to some degree, their rank order of priorities is not sufficiently different from those of other respondents when measured in the aggregate.
Again, what about over-time change?We see a bit more movement in these percentages than for those presented in Figure 3, but none that suggests any common directional change.As a result, we again conclude that the GSS suggests considerable stability in working-class attitudes toward government responsibility in these four domains.
Government responsibility for the safety net.In Figure 5, we present results for an additional four items from the GSS government responsibility module, and these are focused on support for the safety net.GSS respondents were asked the following: Figure 4: Class differences among non-Hispanic whites in opinions on the government's responsibility for addressing inequality and managing the economy in 2006 (black) and 2016 (green).The sample is composed of eligible voters in the 2006 and 2016 GSS who self-identify as non-Hispanic and white only and have an assigned class because they reported a current or last occupation.The numbers of respondents are 1,636 (upper left), 1,659 (upper right), 1,660 (lower left), and 1,642 (bottom right).The variation in sample size reflects different rates of "don't know" and refusals for each item.The marginal prediction for 2016 for the job question could not be computed for classes IVc and VIIb because of a lack of variation in the small proportion of the sample that is in these two classes.
On the whole, do you think it should or should not be the government's responsibility to . . .Provide for a decent standard of living for the unemployed?Provide health care for the sick?Provide a decent standard of living for the old?Provide decent housing for those who can't afford it?Figure 5: Class differences among non-Hispanic whites in opinions on the government's responsibility for the social safety net in 2006 (black) and 2016 (green).The sample is composed of eligible voters in the 2006 and 2016 GSS who self-identify as non-Hispanic and white only and have an assigned class because they reported a current or last occupation.The numbers of respondents are 1,642 (upper left), 1,667 (upper right), 1,666 (lower left), and 1,642 (bottom right).The variation in sample size reflects different rates of "don't know" and refusals for each item.The marginal prediction for 2016 for the housing and old questions could not be computed for classes IVc and VIIb because of a lack of variation in the small proportion of the sample that is in these two classes.
Here, class differences are again present, but they appear less substantial than in Figures 3 and 4. Working-class respondents favor greater government responsibility for providing a "decent" standard of living for the unemployed.Perhaps because of the collective memory (or experience) of the Great Recession, support for this type of government responsibility increased uniformly between 2006 and 2016.Class differences in government responsibility for "health care for the sick" follow the sociological science | www.sociologicalscience.com same weak class pattern.And although government responsibility for health care remains high, it did fall uniformly for all classes between 2006 and 2016 (except for the very imprecisely estimated rate for famers and agricultural workers).We cannot be certain that these small, uniform changes are genuine because of their size relative to expected estimation error.Nonetheless, the second pattern of change is consistent with other results using more GSS respondents and additional years that suggest declining support for government involvement in health care following the passage of the Patient Protection and Affordable Care Act, probably because of the postpassage campaign that the Republican Party waged against it (see Morgan and Kang 2015).
For the item on a "decent standard of living for the old," all classes agreed that the government bears considerable responsibility, and this uniformity may dampen class differences because of a ceiling effect.For the item below it on "decent housing for those who can't afford it," class differences perhaps are a bit more prominent.For both of these last two items, we see very little evidence at all of change over time.
Altogether, these 10 items suggest a common narrative that is consistent with the white working-class narrative.Looking within the pool of non-Hispanic white, eligible voters, members of the working class feel disproportionately powerless over their government and only a little bit less confident in their understanding of political matters.They are more likely to favor an active government in managing the economy, including efforts to reduce inequality, but their relative support for the social safety net is less substantial than one might expect on the basis of their likelihood of needing its protective benefits.Perhaps most importantly for the goals of this article, their attitudes appear to have changed little between 2006 and 2016.When some very modest change might be present-for unemployment support and health care for the sick-the case for change only seems to barely pass a threshold of speculation because of small, uniform change across all GSS respondents, suggesting a "thermostat" type of change that is not itself class related (see Morgan and Kang 2015).
Prejudice, affirmative action, and immigration.The secondary theme of the white working-class narrative is that non-Hispanic whites in the working class are more prone to racial prejudice, anti-immigrant sentiment, and emergent forms of right-wing, populist white nativism.As a result, some portion of the white working class can be mobilized to vote by candidates who appeal to interests of this type.
The GSS offers many items on attitudes that can inform this theme, and for an excellent overview of findings from the racial attitude items available since the 1970s, see Bobo et al. (2012).In Figure 6, we offer results of four representative items, and we analyze them in this article with the same class coding utilized above. 21We first considered items on racial intermarriage, about which GSS respondents were asked: What about having a close relative marry a black person?Would you be very in favor, somewhat in favor, neither in favor nor opposed, somewhat opposed, or very opposed?sociological science | www.sociologicalscience.com Figure 6: Class differences among non-Hispanic whites in racial prejudice, attitudes toward affirmative action, and the level of immigration to the United States from 2004 through 2014 (black) and in 2016 (green).The sample is composed of eligible voters in the 2004 to 2016 GSS who self-identify as non-Hispanic and white only.The numbers of respondents are 6,099 (upper left), 5,899 (upper right), 6,097 (lower left), and 5,539 (bottom right).The variation in sample size reflects different rates of "don't know" and refusals for each item.The marginal prediction for 2016 for the preferential hiring question could not be computed for classes IVc and VIIb because of a lack of variation in the small proportion of the sample that is in these two classes.
What about a Hispanic or Latin American person?Would you be very in favor, somewhat in favor, neither in favor nor opposed, somewhat opposed, or very opposed?
These two items tap forms of racial prejudice grounded in the desire to preserve and promote racial separation.
To enable a consideration of group threat, which is now often referred to as racial resentment, we consider these two items: Some people say that because of past discrimination, blacks should be given preference in hiring and promotion.Others say that such preference in hiring and promotion of blacks is wrong because it discriminates against whites.What about your opinion: are you for or against preferential hiring and promotion of blacks?Do you think the number of immigrants to America nowadays should be increased a lot, increased a little, remain the same as it is, reduced a little, or reduced a lot?
The first item allows for the construction of an outcome that expresses opposition to affirmative action in the workplace but likely also opposition to affirmative action in other domains as well.The second item allows for the construction of an outcome that expresses anti-immigrant sentiment.
Figure 6 shows that all groups of non-Hispanic whites displayed some degree of racial prejudice and susceptibility to group threat, belying lay claims that survey respondents are unwilling to offer responses that are indicative of racial animus.Before considering class differences, we should note that these GSS items were asked in each biennial survey for the GSS from 2004 through 2016.Accordingly, we have a good deal more information on which to base these estimates than for the analysis above in Figures 3 through 5.In particular, we are able to offer a comparatively precise modeled estimate for each item for 2004 through 2014 and then consider whether an analogous estimate for 2016 offers any evidence of change. 22Notice that the error bar for each item for 2004 through 2014 (in black) is considerably shorter than the error bar for 2016 (in green).Now, to carry on to some sobering substance, consider first prejudice in the form of a preference for racial separation as measured by opposition to "a close relative marrying a black person" and "a close relative marrying a Hispanic person."Even for our comparison group of white-collar workers, about 20 percent would oppose a close relative marrying a black person, and more than 10 percent would oppose marrying a Hispanic person.For the working class, the corresponding rates of opposition are higher at about 30 percent and 20 percent, respectively.In addition, we see almost no evidence of change in 2016 in comparison with the 2004 through 2014 estimate; the point estimates are almost exactly the same for the working class across both time periods.
The upper-right panel demonstrates that non-Hispanic whites strongly oppose affirmative action in hiring and promotion, regardless of class position.Although opposition may have declined a small amount in 2016, the change is too small to yield any clear conclusion because of sampling error.And, even if genuine, the change seems trivial relative to the overwhelming opposition to this type of affirmative action.
Finally, the lower-right panel indicates that a clear majority of non-Hispanic whites favor at least some reduction in immigration.As with racial prejudice, this anti-immigrant sentiment is more substantial for the working class, with about 60 percent supporting at least some reduction in immigration.We cannot conclude from these results or from any other GSS items whether the preference for a reduction in immigration applies to immigrants from all origin countries or only some countries and whether it applies across the full range of the education and skill distribution of immigrants or not.Nonetheless, a clear majority of the working class favors a reduction in immigration, and the preference, on average, did not change between 2004 and 2016.
Altogether, the results in Figure 6 suggest that racial prejudice among non-Hispanic whites remains common, and, furthermore, it is more common among working-class, eligible voters than others. 23It is difficult to imagine a scenario in which racial prejudice is not an important component of both opposition to affirmative action and anti-immigrant sentiment.The GSS cannot reveal in any clear fashion how much of the preference for reduced immigration is attributable to prejudice and related forms of nativism or instead to a perceived group threat to economic security.The two are intertwined in the survey responses and nearly certainly within the minds of many non-Hispanic white, eligible voters.Few voters can cleanly apportion within their own minds the multiple sources of support for a particular policy or candidate, especially given the dynamic oscillation of components of sentiment.It would be odd to expect white, working-class voters to be any better at this complex cognitive task than others.
What we do not have evidence for in the results of Figure 6 is any substantial change over time on these items.Most of the 2016 GSS interviews occurred from late spring through late summer, and by that time the dominant themes of the presidential election campaign were set.Accordingly, the GSS suggests remarkable stability during a period of rising racial tension and a decline of generalized comity.In other words, the GSS does not support the claim that the racial prejudice or anti-immigrant sentiment of non-Hispanic whites increased in response to the political climate through the first half of 2016.But, as we discuss below, it was high enough already that an opportunistic candidate willing to break with the norms of established political discourse on intergroup relations could exploit it to his advantage.

Turnout Rates and Compositional Change from 2012 to 2016
We now return to turnout patterns, having established two key findings that are consistent with the white working-class narrative.In our analysis of the CPS-VRS in the first section of results, we showed that the turnout rate increased relatively more for working-class whites.Coupled with the GSS results just offered, it seems reasonable to infer that the relative turnout rate increase was the result of the particular appeal of the Trump campaign.
To complete our analysis, we offer in this section a brief consideration of what the CPS-VRS implies about the changing composition of the voters who turned out to cast ballots in both 2012 and 2016. 24Class-and race-specific turnout rates are one source of compositional change, but so are shifts in the underlying distributions of race-ethnicity and class.We first introduce the evolving distribution of raceethnicity into the analysis.
Table 3 presents the percentages of each class and education group that selfidentified as non-Hispanic white and voted in 2012 and 2016 in competitive states.In other words, unlike the turnout rate analysis presented in Figures 1 and 2, which implicitly held our two-category distribution of race-ethnicity fixed, in Table 3 we allowed the marginal shift in race-ethnicity between 2012 and 2016 to enter into the results and interact with the turnout rate.
The results in Table 3 show that an underlying increase in the relative size of the pool of eligible voters who were nonwhite or Hispanic mitigated the consequences of the turnout rate changes presented above in Figures 1 and 2. In particular, the share of the working class that was non-Hispanic, white only, and voted increased only from 36.2 percent to 36.7 percent in competitive states.Thus, the increasing race gap in the turnout rate, which grew to about 7 percent in Figure 1, resulted in a much smaller relative increase in white, working-class voters because of an underlying compositional shift in race-ethnicity within the working class.It seems likely that the mitigation provided by demographic change was less substantial in blue wall states, where population growth is lower than in other competitive states, but the CPS-VRS lacks the sample size to permit a state-level analysis when disaggregated by class.
For completeness, note also that demographic change is relevant for all other classes and education groups as well.For example, we showed in Figure 1 for white-collar workers that the turnout rate increased a bit among non-Hispanic whites and declined substantially among all others.Table 3 shows that, nonetheless, the percentage of white-collar workers who were non-Hispanic white and voted decreased from 63.2 to 61.1 percent between 2012 and 2016.Changes such as these are precisely those that support the position that demographic forces favor political parties that can appeal to a broad multicultural electorate.
Extending the comparison in Table 3, Table 4 allows for the marginal distributions of class in 2012 and 2016 to structure the compositional patterns as well.Table 4 reports the percentage representation of 14 separate groups in each election year: the seven class and education groups by the two categories of race-ethnicity.
To clarify the calculations in the table, consider farmers and agricultural workers.Figure 1 shows that the turnout rate among non-Hispanic whites in this group increased substantially, and Table 3 shows that the marginal shift in race-ethnicity did not erode the apparent relevance of this change because the percentage in Table 3 increased by almost 10 percent between 2012 and 2016.However, as Table 4 shows, farmers and agricultural workers are such a small portion of the population that the increase in the turnout rate resulted in a gross representational increase of only about one-tenth of one percent for non-Hispanic white farmers and agricultural workers (i.e., from 0.6 percent to 0.7 percent).Now consider the other class and education groups in Table 4, all of which show a common pattern.Even though the turnout rate increased for most groups of non-Hispanic whites relative to all others, the gross representation of non-Hispanic white voters did not increase for most groups because of compositional shifts in both the class distribution and the distribution of race-ethnicity.Most importantly, the white working class constituted 12.8 percent of voters in 2012 but only 12.2 percent of voters in 2016.In other words, the size of the white working class declined in size between 2012 and 2016, even as economic conditions improved in the country, and this marginal shift more than counterbalanced the turnout rate increase shown in Figures 1 and 2.
We see two important implications of these final results beyond their obvious descriptive utility.First, a comparison of Figures 1 and 2 with Tables 3 and 4 justifies the need to analyze turnout rates directly if one is genuinely interested in whether individual voting propensities changed for different groups of eligible voters between 2012 and 2016.If we had only tabulated the results in Table 4 and had not otherwise presented direct estimates of turnout rates in Figures 1 and 2, then one would be tempted to conclude that turnout rates did not change and that the case for the unique appeal of the Trump campaign to white, working-class voters had less support than the New York Times claimed just after the election.This seems to be an interpretive strategy favored by some data journalists inclined to doubt the validity of the white working-class narrative.
Second, for those looking for a silver lining in turnout patterns, Table 4 should be reassuring.Class-specific turnout rates, although consistent with the white working-class narrative, must also contend with compositional changes produced by demographic momentum.The relative turnout rate increase among white, working-class voters may have been necessary to push Trump just over the threshold of victory, but the surge itself was too small in gross terms to generate a more decisive victory in the closest states.

Conclusions
In the first portion of our analysis, we showed that the CPS-VRS provides some support for the white working-class narrative for the Trump victory in 2016.The turnout rate of non-Hispanic, working-class whites in competitive states did increase in 2016, and the impact of this increase was enhanced by a simultaneous relative decrease in the turnout rate of other eligible voters within the working class.As a result, the race-ethnic gap measured in this way was substantially larger in 2016 than it was in both 2008 and 2012, when the Republican candidates lost.This is clear evidence of a relative surge among non-Hispanic, working-class whites even if the magnitude of the turnout rate surge may be only a few percentage points.This shift may have been particularly consequential because it was concentrated in the blue wall states of Pennsylvania, Michigan, and Wisconsin, but the CPS-VRS lacks the sample size to detect such localized change.
Consistent with the county-based analysis of votes discussed in the introduction, the turnout rate increase was even more substantial for non-Hispanic whites who worked as farmers, ranchers, or other agricultural workers.For the class voting literature, farmers and agricultural workers are not usually considered members of the working class, and so this pattern does not support a white working-class narrative that academic researchers would have constructed.Whether it is supportive of the working-class narrative splashed across the front page of the New York Times just after the election is probably a matter of reader interpretation.
The GSS analysis suggests that these additional white, working-class voters could be expected to favor the material interests of the working class and support appeals to subvert multiculturalism.As a result, we cannot make any convincing case that the modest turnout rate surge of white, working-class voters in 2016 is attributable to either the "working class" or the "white" portion of the group identity that appears to be a source of the relative turnout rate increase.Indeed, many working-class voters mobilized by the Trump campaign are unlikely to be able to apportion their enthusiasm between policies and rhetoric that promote working-class economic interests and those that supposedly facilitate a return to a past when a less multicultural United States had a more prominent and secure position in the world economy.

Discussion
Conventional wisdom is sometimes wrong.The more common problem is that it is poorly calibrated.The white working-class narrative for the 2016 election outcome is a typical example.It was constructed quickly to account for an unanticipated event by using largely suggestive evidence from a long campaign and then supplemented just after the election with exit poll data that contained no direct measure of its key actor: the white working class.
Our overall goal is to calibrate this conventional wisdom and thereby begin to evaluate it more deeply.In our analysis in this article, we have used two data sources, the CPS-VRS and the GSS, with which we can deploy a genuine direct measure of the working class and build on decades of prior research and debates on class politics. 25On balance, as summarized above, we see considerable support for the conventional wisdom and yet many unresolved questions that demand further attention.
The most important unresolved question is whether a direct analysis of the relationship between class position and votes cast in 2016 will line up with the white working-class narrative.So far, social science has collectively failed to generate any current data source that delivers both data on class and data on votes cast that can be analyzed now.With the collection of the 2018 GSS, we will have retrospective self-reports of votes cast (along with the errors typical of voting data when recalled 18 months after the fact).When an analysis of the relationship between class and votes cast in 2016 becomes possible, it may be that Obama-to-Trump voters, possibly among dealigning independents (see Morgan and Lee 2017), will be a much more important component of the full explanation of the 2016 election outcome than the modest turnout rate surge that we show in this article.
Even for our more limited analysis, further work is needed to develop answers for some lingering questions.The relative turnout rate surge among non-Hispanic whites in competitive states was not dramatic.Nonetheless, it is notable, we think, that it emerged in a period when the working class was moving away to some degree from traditional identification with either the Democratic Party or the Republican Party, especially among those who had not recently voted (see our analysis in Morgan and Lee 2017).More work on turnout, using measures of party identification, will be needed to probe these relationships, and this also will be possible with the release of the 2018 GSS data.
Looking toward future elections, and with the recognition that the CPS-VRS overstates turnout rate differences between highly educated voters and others to some degree, a massive gap in turnout nonetheless exists between white-collar workers and members of the working class.An important implication of this pattern is that there are many more individuals who can be converted from nonvoters to voters within the working class than among white-collar workers.These nonvoters in the working class represent an untapped source of political power and are not obviously aligned with either party.If working-class voters who are white and non-Hispanic can be brought into voting booths at increasing rates in 2018 and 2020, then populist white nativism may have more longstanding support than many commentators assume.Then again, as our compositional analysis showed, basic demographic change may be able to effectively contain it, assuming no further turnout rate declines occur among nonwhite and Hispanic eligible voters.
Our analysis of the GSS suggests that contrary to concerns that a new equilibrium of racial resentment has arisen, rather little has changed between 2004 and 2016 for non-Hispanic whites.Of course, the most recent GSS data were collected in the middle of 2016, and much has transpired since then.Our conclusion so far has been this: if Trump's 2016 victory is at least partly attributable to the racial resentment of the white working class, and possibly other white voters, it is an opportunity seized and effectively exploited rather than one that Trump himself created during his campaign.This conclusion may need to be revised as the dynamic post-2016 political environment evolves.It is certainly possible that a wider examination of the GSS will support more change in racial attitudes than our analysis in this article has revealed.It could be, for example, that the small uptick in prejudice and anti-immigrant sentiment, on average, for members of classes IVab and V (see Figure 6) is the beginning of a trend that will need to be investigated fully when the 2018 GSS is returned from the field.
Finally, looking back at the CPS-VRS turnout rate results, one pattern deserves additional scrutiny.By our reading of the evidence, turnout rate patterns were most similar in 2004 and 2016, with working-class whites turning out at higher rates in both years.It may be that we should be interpreting the 2016 election in light of this similarity.Whereas post-9/11 patriotic devotion propelled Bush to a reelection victory in 2004 with a coalition that included white, working-class voters in swing states, in 2016 it would appear that an appeal to white, populist nativism was similarly effective with many of the same voters.If this similarity has interpretive value, then the seeds of dealignment with the establishment Republican Party may have been planted during the Bush presidency.The elections in 2004 and 2016 can then be reconciled, and the rise of the Tea Party movement during the Obama presidency can be seen as the necessary catalyst that transformed a prior patriotic devotion into the right-wing, populist fury that just barely secured victory for Trump.eligible voters is first identified and after which nonresponse patterns are tabulated in order to understand participation.Instead, the exit poll is a clustered, opt-in design.
4 The ANES collects the information but does not release codes that can be matched to the full distribution of occupations.Even a coarse coding of occupation has not been released by the ANES since 2004.The other surveys and polls do not even ask for the information.
7 A third component of the explanation is also sometimes discussed.For surveys and polls that are focused on political matters, respondents feel pressure to claim that they voted so that others will regard their responses to attitudinal questions as meaningful.This overreporting is a type of social desirability bias and is likely more prominent among eligible nonvoters with higher levels of education.
8 Although the CPS overestimates turnout as well (by about 10 percent), it does not do so to the degree that most other data sources do.Its sampling design also allows it to be adjusted, as we explain in the online supplement, by known state vote totals.And we know of no evidence that the unadjusted upward bias varies substantially over the years we consider here in a way that would compromise our decomposition of turnout rates.
14 The CPS-VRS asks occupational and related questions only of sample members who are currently (or were recently) in the labor force.As discussed below, the GSS asks for the most recent occupation of all sample members, generating a class position for 98 percent of respondents in recent years.
19 As explained in the online supplement, we excluded about 2 percent of GSS respondents who reported that they had never held an occupation of any type.In addition, we excluded respondents aged 18 to 24 because their class locations were misleading and because the GSS does not sample college students who live in dorms (or others in group quarters).
20 In particular, the results in Figure 3 are marginal predictions across each year from pooled logit models that fit five coefficients for six age categories (in addition to main effects for year and class as well as all interactions between year and class).The agree-disagree scale was dichotomized so that "agree" represents the response option of both "agree" and "strongly agree." 21 Scales of racial prejudice can be constructed from items such as those we considered and additional ones, and we offer results in the online supplement that do so.The results presented there are consistent with the analysis offered in the main text.
22 The results in Figure 6 are marginal predictions from pooled logit models that fit five coefficients for six age categories, an interval-scaled term for year, main effects for class group interacted with the term for year, and a dummy variable for 2016 that is also interacted with class group.The point estimate for each class group for 2004 through 2014 is the marginal effect for each class group across years (a smoothed average, having removed the discrete shift for 2016 with its year-specific dummy, which can be thought of as an estimate in the middle of the time series, which is 2009, or the first year of the Obama presidency).The point estimate for each class group in 2016 is the marginal effect for 2016, which adds a modeled percentage generated by the class-specific 2016 difference to the underlying year effect generated by the linear term for year.
23 Still, as shown in Figure 6, there may be some evidence that prejudice is even more pronounced among farmers and agricultural workers, but estimation error is large.In addition, there may be some subtle evidence of an increase in prejudice among the intermediate-class group (IVab and V), which includes nonprofessional self-employed workers as well as a large proportion of public safety workers.Again, however, estimation error is substantial.When the 2018 GSS data are collected and released, these small changes will be worthy of considered attention.
sociological science | www.sociologicalscience.com 24 A full consideration of compositional change is beyond the scope of this article, and it would need to attend to a variety of complications.Chief among these are (1) adjustments for relative upward bias in self-reported voting by more highly educated respondents; (2) adjustments for selection into the labor force, and hence a class position in the CPS-VRS; and (3) a more fine-grained analysis of the residual education groups, separated into those retired, on disability, in school full-or part-time, or never in the labor force at any point.
25 Some prominent data journalists have been rather myopic on these issues.Nate Silver (2016), for example, wrote in his county-based analysis: "Are these so-called white working-class counties?You could argue for it: They're mostly white, and they have average or below-average incomes.But, of course, 'class' is a slippery term, and definitions vary."After his analysis comparing education to income at the county level, he concluded: "In short, it appears as though educational levels are the critical factor in predicting shifts in the vote between 2012 and 2016."We fail to see how a closer relationship to education than income can be regarded so easily as evidence against the importance of class, which appears to be Silver's conclusion.

1
Classes I, II, and IIIa (white-collar group) Classes IIIb, VI, and VIIa (working-class group) Classes IVab and V (intermediate group) Classes IVc and VIIb (farmers and agricultural workers)

Figure 1 :
Figure 1: Class differences in voter turnout rates in 18 competitive states, 2004 to 2016.Non-Hispanic whites are shown in red, and all others are shown in gray.
sociological science | www.sociologicalscience.com 2 Bachelor's degree or more Some college High school diploma or less

Figure 2 :
Figure 2: Differences by education group in voter turnout rates in 18 competitive states among respondents not currently in the labor force, 2004 to 2016.Non-Hispanic whites are shown in red, and all others are shown in gray.

Figure 3 Figure 3 :
Figure 3 presents the percentage agreement with these two statements separately by class groups and by year.The dots are point estimates of percentages, and the lines are plus-or-minus one standard error (truncated at the bound of 100 in

Notes1
When seeking more evidence, many long-form journalists and opinion writers sought corroborating evidence that could be extracted from media-organized focus groups as well as book-length testimonials-usually either Vance's 2016 memoir, Hillbilly Elegy: A Memoir of a Family and Culture in Crisis, or Hochschild's 2016 academic book, Strangers in Their Own Land: Anger and Mourning on the American Right.The more learned among these writers also sought support in Cramer's The Politics of Resentment: Rural Consciousness in Wisconsin and the Rise of Scott Walker.All three books provide evidence for the narrative, albeit somewhat indirect because of their disproportionate attention to whites who live in rural areas.

Table 1 :
The class schema utilized for both the CPS-VRS and GSS analyses.

Table 2 :
Class and education group by a two-category coding of race-ethnicity in competitive states in 2016.

Table 3 :
Percentages of each class or education group that self-identified as non-Hispanic, white only and voted.Notes: CPS-VRS sample members in competitive states in 2012 and 2016 without missing data on voting questions, weighted by using the scaled weights explained in the online supplement.The raw N for the table is 63,996.

Table 4 :
Percentage representation of 14 groups of voters defined by class and education group and by self-identification as non-Hispanic and white only.Notes: CPS-VRS sample members who voted in competitive states in 2012 and 2016, weighted by using the scaled weights explained in the online supplement.The raw N for the table is 47,542.
The group Nonprofit VOTE reports a turnout rate among eligible voters of 60.2 percent (see Pillsbury and Johannesen 2017) based on the United States Election Project maintained by Michael McDonald (see http://www.electproject.org).6 See, for example, Holbrook and Krosnick (2010) and