The Politics of Losing

by Rory McVeigh

  The geographic diffusion of the Klan in the 1920s was impressive considering its association with the original Klan, famous for its violent oppression of black Americans in the postwar South. The Imperial Night-Hawk, the Klan’s national newspaper, regularly listed events—rallies, speeches, chapter foundings, and charitable activity, like plans to build a new hospital in Topeka, Kansas6—to demonstrate to the Klan faithful that they were part of a powerful movement, one whose influence extended across the nation. While many of these events were reported in Texas and Georgia (the Klan was headquartered in Atlanta and Imperial Wizard Hiram Evans hailed from Dallas), other Southern states reported relatively little activity. Over two years, the Night-Hawk listed only 22 events for North Carolina, 29 for Virginia, and 58 in Mississippi. Midwestern states, on the other hand, were hotbeds of Klan activity: 158 events in Illinois, 168 in Indiana, and 213 in Pennsylvania.

  FIGURE 5.1 Number of Klan events reported in the Imperial Night-Hawk by state, 1923–1924.

  Despite the Klan’s far-reaching influence, not all communities were friendly to it. Klansmen were often on the receiving end of violence, for example, when they ventured into enemy territory that put them in contact with manufacturing laborers or concentrations of Catholics and immigrants. In the summer of 1923, a Klan parade came under attack in Carnegie, Pennsylvania.7 As night fell, Klansmen lit a firework display and a sixty-foot burning cross before marching toward Carnegie. Their arrival triggered a riot, in which Thomas Abbot, a new Klan recruit, was fatally shot.8 In 1924, Klansmen attempted to march into South Bend, Indiana, only to be beaten back by Catholic students from the University of Notre Dame. “As the Klansmen left the hall,” one wrote, “they were pounced upon, beaten, and cursed by the students of Notre Dame. The Klansmen, as is their custom, refrained from fighting back those who opposed their movements and actions, again proving to the world that they are law-abiding citizens, willing and ready to let the law take its course.”9 Later in the same article: “The various attacks over the country against members of the Knights of the Ku Klux Klan are only a demonstration of the un-American interests against anything that is American and of Protestant extraction.”

  There are patterns to Klan mobilization, which we can see by identifying the types of communities that should have been receptive to the Klan’s overtures.10 Prior research shows, for example, the Klan was particularly active in states where manufacturers had expanded and hired more workers, part of the increasing implementation of the sort of industrial mass production that Klansmen resisted.

  States that gained the fewest new voters from women’s suffrage were similarly ripe for Klan activity. This finding seems counterintuitive, but it makes sense when recognizing the Klan as a national movement engaged in national conflicts. New voters were distributed unevenly across states because several had already extended the vote to women before the Nineteenth Amendment passed. Many of these new voters were concentrated in northeastern industrial states, where the Klan’s enemies were also concentrated. The total number of votes cast in New York, for example, increased by 70 percent from 1916 to 1920. In Massachusetts, the increase was 86 percent. In the state of Washington, there was only a modest increase of 4 percent, and in Illinois, the number of votes actually declined by 4 percent. Klansmen had extra incentive to organize women voters in these states, so that politicians would not ignore them while focusing on voters in northeastern states.11

  Looking at Klan membership in the counties of Indiana, there is strong evidence that the Klan enjoyed its most successful recruiting in communities whose residents would have responded to the Klan’s framing of lost power—for example, in counties experiencing industrial concentration, but also in agricultural economies that produced farm goods, since agricultural exports to Europe plummeted after the war.12

  IDENTIFYING TRUMP STRONGHOLDS

  In chapter 4 we explored disappearing economic power and how Trump’s brand of white nationalism had an intuitive appeal to Americans in communities passed over by the global economy. Trump’s strong support in those places secured him the Republican nomination. Afterward, traditional Republican voters mostly fell in line for the general election.

  If we looked only at results of the general election, we might think all of our discussion of power devaluation and the disruption of intraparty alliances is misguided. When we examine variation across counties in the general election vote for Trump and the general election vote for Republican Mitt Romney four years earlier in 2012, the correlation between the two is very strong. Each dot on the graph represents a county in the United States. The horizontal axis shows the percent vote for Romney and the vertical axis shows the percent vote for Trump. With just a handful of exceptions, Trump counties were also Romney counties, and Trump fared poorly in counties where Romney also struggled.

  But the general election obscures Trump’s impact on Republican politics. To see how he disrupted the party, we must first look at where he received support in the primaries, when he was pitted against other Republicans. Before presenting our analysis, however, we zoom in on a few cases that represent the sorts of communities that underlie the broader statistical patterns.

  FIGURE 5.2 Percentage of votes for Romney in 2012 versus percentage of votes for Trump in 2016 by county.

  Source: Data from “United States Presidential Election Results,” Dave Leip’s Atlas of U.S. Presidential Elections.

  North Carolina was a key swing state in the election, and Democrats and political prognosticators expected that Clinton would win there. On Election Day, statistician Nate Silver’s FiveThirtyEight website gave Clinton a 55 percent chance of coming out on top.13 Trump, however, won North Carolina with 49.8 percent of the vote, compared to Clinton’s 46.2 percent.

  Three counties in the center of the state—Wake, Durham, and Orange—make up what is called the Research Triangle. Each has a major university—North Carolina State, Duke, and the University of North Carolina at Chapel Hill—that prepares graduates to fill high-tech occupations in the global economy. Based on census data collected from 2010 to 2014, in each county the number of residents ages twenty-five and over with at least a bachelor’s degree was more than 45 percent of the population. This highly educated labor force has attracted floods of good jobs to the region for decades. Research Triangle Park was developed in 1959 to prevent graduates from the region’s top universities from leaving the state. The park started as an expansion of IBM, but now houses 170 companies, like Biogen Idec, Syngenta, United Therapeutics, Bayer Crop Science, Eisai, BASF, the EPA, and the National Institute of Health’s National Institute of Environmental Health Sciences.14

  All three of these counties vote Democratic, and Republican primary voters in the counties were by no means enamored with Trump in 2016. Trump carried the overall state with 40.2 percent of the vote, while Texas senator Ted Cruz came in second with 36.8 percent. In the Research Triangle counties, however, Trump lost to Cruz by a significant margin. In the general election, Trump secured only 37.2 percent of the vote in Wake, 22.5 percent in Orange, and 18.6 percent in Durham.

  In “this” North Carolina, Trump had little appeal to either Democrats or to Republicans. But in the “other” Carolina, he was quite popular. Consider Graham, a small county on the western edge of the state, bordering Tennessee, with a population of about nine thousand. Just under 90 percent of the population is white, and less than 1 percent is black. Only 16.6 percent of the over-twenty-five population are college graduates. More striking, 20.2 percent did not complete high school. Median household income is only $37,000. The unemployment rate is not unusually high (4.9 percent) but, more tellingly, the number of residents ages sixteen and older not in the labor force is a staggering 51.4 percent.15

  In 2014, the Citizen Times, of nearby Asheville, North Carolina, ran a feature story about Graham County titled “When the Last Factory Leaves a Mountain Town.” The last factory in this case was Stanley Furniture, which had just announced that it was closing the last manu
facturing plant in Robbinsville, North Carolina, laying off four hundred workers in a town where only 3,800 people were employed. “Stanley and Robbinsville,” declared the Citizen Times, “were only the latest casualties in a generation-long decline of manufacturing in the mountains. The eighteen counties of Western North Carolina had 61,344 factory workers in 1990. By 2013, the region had lost 58 percent of those jobs with only 25,580 men and women drawing a manufacturing paycheck.” Said one resident, “If somebody doesn’t open it back up and getting some jobs in there, I’m afraid this town will kind of die off. I would think about leaving too.”16

  Despite the economic blight, Graham County has consistently voted Republican in presidential politics. Since George H. W. Bush, every Republican candidate has averaged about 68 percent of the Graham vote in the general election. Then came Donald Trump, who promised to address the economic circumstances in struggling white communities like theirs. Trump picked up 40.2 percent of the primary vote statewide and 52 percent of Graham County. In the general, he received 78.8 percent of the vote.

  VOTING TRENDS

  With these cases in mind, let’s examine patterns in the primary votes and the communities where Trump attracted strong support in the nation as a whole. Was college education vital to determining whether Trump’s message resonated with voters or repulsed them? While Clinton fared better than Trump among those with college degrees, Trump received slightly more of the share of white voters with a college degree.17 While we already know that educational difference correlates with the Trump vote, we have argued so far that we should focus on the community more than the individual. Local economies must have a critical mass of college degrees if they are to connect to the global high-tech economy. And communities with highly educated populations reaped most of the gains of the postrecession recovery. We expect that Trump’s message, therefore, had the greatest appeal where economies stagnated, where there were too few well-paying jobs for those without advanced degrees. Even college graduates in these communities would face hard times.

  To analyze attributes of counties that were more (and less) supportive of Trump in the primary and caucus votes, we use a statistical tool called ordinary least squares regression. There is a detailed and (when necessary) technical account of our methodology in the appendix. Regression analysis estimates the effect that a particular community’s attributes had on the vote for Trump, while taking into consideration (or statistically holding constant) how other county attributes account for variation in that support. For example, we expect that Trump enjoyed very strong support in counties where a relatively small proportion of the adult population held a college degree. However, we also know that counties with relatively low proportions of college graduates tend to have high proportions of evangelical Protestants.18 To obtain a good estimate of the effect of only education on the vote, we must factor out the effects attributable to religion.

  We can examine the vast majority of counties in the United States: almost 2,900 out of 3,142, or about 92 percent of all counties.19 Because our argument has to do primarily with how features of local communities influence the Trump vote, we statistically control for differences between states so that our results are limited to only the effects of county attributes. This approach controls for differences in the election processes of the different states during the party nomination process. For example, the caucus process is very different from an open primary, and later state elections featured fewer candidates than did the earlier ones. We present the full results of our regression analyses in the appendix. Here, we describe the most relevant findings.

  What did we find? First, the population density of counties and the percent that voted for Mitt Romney in the 2012 general election do not predict the vote for Trump. Median household income also did not predict the vote for Trump. Trump tended to secure more votes in counties with lower incomes, but that relationship does not hold up when including other variables that predict variation in the voting outcomes. The most important of these variables is the percent of county residents over the age of twenty-five who hold a college degree. This strongly predicts the voting outcome, even after controlling for other factors. The coefficient for the education variable is negative, which means that the more college graduates, the lower the vote for Trump tended to be. Controlling for other factors, an increase of just 1 percent of college graduates in a county reduces the Trump vote by about .3 percent. If we were to compare a county in which 25 percent of adults held a college degree to one where only 15 percent did, the vote for Trump would be, on average, approximately 3 percent lower in the county with more graduates.

  This is no surprise. Trump’s message should have resonated most strongly in communities where few residents had the credentials to thrive in a global economy. He tended to gain strong support in counties with relatively high unemployment rates. He also won counties with high median ages, perhaps reflecting the difficulty older Americans have adjusting to the changes of postrecession America. In general, these findings indicate that Trump distinguished himself from his Republican competitors by winning more support in counties with few college graduates and where people faced economic hardships.

  Trump’s appeal was rooted at the intersection of economics and what his supporters thought about gender, religion, and race. The more traditional gender arrangements in a county, the more support for Trump. For example, he tended to receive a smaller share of the vote in counties where higher percentages of women worked. He also received less support in counties where college graduates in the county were disproportionately male. This likely reflects his appeal in counties where men were less educated and looked to Trump, hoping he would bring good jobs that didn’t require a college degree to their communities. Communities with ailing economies may have fewer marriages, as couples (and men in particular) may not have the money or stability to settle into durable relationships. Trump won more votes in counties where the percent of adults who were married is low, but that finding falls just shy of statistical significance.20 Yet, as we show in the following pages, the link between voting for Trump and marriage rates differed depending on education. Taken together, these findings suggest that Trump was particularly appealing to voters where the traditional “male as breadwinner” family prevailed.

  Even though Trump himself does not appear to be religious, and compared to many of his competitors he invested little effort in appealing to religious conservatives, he surpassed his Republican opponents in counties with high proportions of evangelical Protestants and in counties with high proportions of Catholics. We explore Trump’s appeal to conservative Christians in more depth in chapter 7 when we consider social status. For now, though, it is worth noting that Trump bucked expectations by beating his Republican opponents in evangelical strongholds.

  And, surprisingly, the nonwhite percentage of a county was not a predictor of the Trump vote. Direct competition with minorities over jobs was apparently not a primary motivator for Trump’s overwhelmingly white supporters. Instead, Trump framed their economic woes in terms of foreign competition and the dearth of jobs at home. Yet we suspect that Trump’s bald appeals to white voters connected with Republican voters in some communities, and especially made a difference in the general election, which we demonstrate later in this chapter.

  Trump fared better than his Republican opposition in counties where a high proportion of the labor force worked in retail. This is the new dual economy at work, in which economic transitions have pitted occupations that require skill, training, and education, and provide relatively high levels of compensation, job security, and opportunities for promotion against those that require little training and pay poorly, mostly in the service sector. Starting in the 1970s, retail jobs began to replace the more lucrative work in manufacturing and mining.21 Trump was particularly appealing in counties with high proportions of retail jobs, presumably because of his promises to bring well-paying manufacturing jobs back to the heartland. On the other hand, counties that still had
high rates of workers in manufacturing jobs supported Trump less. This was perhaps because manufacturing enterprises operating in the United States depend on global markets for production, making Trump’s strong protectionist agenda unappealing. The effect was modest, however. On average, the vote for Trump would be a little more than 1 percent higher in a county with 10 percent of workers employed in manufacturing compared to a county with 20 percent employed in manufacturing.

  THE EDUCATIONAL DIVIDE

  The more college graduates in a county, the smaller the vote for Trump. This result, based on our regression analysis, is depicted in figure 5.3. This strong education effect is not simply attributable to those without college education finding his white nationalist agenda and overt expressions of bigotry appealing. Instead, we look at attributes of places alongside the attributes of individuals. The number of college graduates in a community affects whether the local economy connects to the global economy. Given the economic transitions of recent decades, communities with few college graduates would have difficulty attracting high-paying jobs and would leave many residents vulnerable to Trump’s promises to reverse their economic fortunes by bringing jobs for unskilled workers back to the United States.

  If we are right about the educational divide, then we should expect to find that the effect of other county attributes on Trump votes depends on the county’s proportion of college graduates. Local unemployment, for example, would be most likely to lead voters to embrace Trump in counties that also had few college graduates. We can test these kinds of arguments statistically and, indeed, that is what we find. In figure 5.4, after considering other county attributes, we see that counties with high unemployment were much more likely to choose Trump over his Republican opponents for the nomination—but that was only true in those counties with relatively few college graduates. In counties where 10 percent of the population had a college degree, higher levels of unemployment led to very strong support for Trump. But in counties where 30 percent of the adult population held college degrees, the unemployment rate had no effect on support for Trump. His proposed remedies for joblessness, in other words, had little appeal in communities with a highly educated work force. Economic protectionism, in such communities, would only exacerbate unemployment. His message was well received, however, in counties with high unemployment and few college graduates.