Before we turn to the findings of the literature review, we must first address a pertinent knowledge gap within the extant body of literature. This is that there is no “White working-class” variable for researchers to code. We need such a variable so that researchers are able to analyse the White working-class in the context of Trump’s victory. Part of the problem stems from the fact that “gold-standard” national benchmark survey’s such as the General Social Survey (GSS) and the American National Election Studies (ANES) do not contain a White working-class variable. Instead, the White working-class are defined by multiple independent variables across studies. This would make sense, given that class is a multidimensional construct. As will be apparent, this is reflected in the fact that author of the literature use variables of levels of income, educational attainment, as well as race, to define the sociodemographic group. The following section elucidates the three ways in which working-class whiteness is defined within the academic literature. Afterward, the Author speculates on the relative weight of these individual factors in the formulation of a working-class White variable by using multinomial logistic regression to model the determinates of working-class whiteness.
2.2. Theoretical Background
Scholars examining White working-class support for Trump during the 2016 primary reason and in the immediacy of the election result initially used measurements of income in relation to the sociodemographic group. In attempting to dispel the “myth” of Trump’s White working-class support, for example, Silver [
22] pointed to state-level exit poll data from Republican primary states, which showed that the median household income of Trump voters was
$72,000, ’well above’ the national median of
$56,000. There were, however, reasons to be wary of the conclusions drawn from data released in the immediacy of the election. This is because Roediger notes that class is ‘not well studied by anyone via instant analysis of election results’ [
23]. Statisticians such as Silver [
24] corrected these conclusions when the exit poll data was released. After the election, Silver [
24] reran the numbers and found that ‘education levels [were] the critical factor’ in determining shifts in the White vote between 2012 and 2016.
Another way that authors of the literature have defined the White working-class is by educational attainment. One’s class is ‘not just about money’, notes Williams [
3] (p. 12). Levels of educational attainment also determine one’s class status. The social and familial networks of American college graduates are entirely different from those of non-college-educated individuals. These differences are best expressed through what sociologist’s call “professional” and “clique” ‘networks’ [
3] (pp. 35–36). College graduates enter professional vocations, forming professional networks. Professional networks are composed of large matrices of acquaintances whom elite professional encounter in their specialised career field. Conversely, the working-class live their lives in tightly-formed and deeply rooted “clique” networks [
25,
26]. These networks have material benefits in working-class communities—people “have each other’s backs”, from babysitting their friend’s children to assisting with house repairs.
However, educational attainment is not the sole determinant of class. Since class is a multidimensional concept, it is better conceptualised using multiple indicators. Reeves et al. [
27] argue that education should be used to supplement rather than substitute other measures of class. We see this consideration in Williams and Boushey’s [
28] report on the work-family conflict in the United States. Here, they define the ‘missing middle’ as those households lying between the bottom third and top fifth percentiles of household income, as well as households in the top percentile in which no adult possesses a college degree [
28] (p. 32). Williams and Boushey’s [
28] definition of class is more inclusive than one based solely on measurements of income because it intentionally includes sections of the American populace who, if defined by income alone, would otherwise be excluded from the middle-class.
Definitions of class in the studies discussed thus far have relied on objective and statistically measurable indicators such as income and education. However, they disregard the importance of an intangible, cultural aspect essential to class definitions if we are to understand its complex multidimensionality fully. This is a categorisation that scholars call subjective class identification [
29]. Subjective class identity is an essential notion in the literature on social stratification. Despite the central role that the notion plays in the literature, however, issues surrounding the measurement and validity of such subjective phenomena raise methodological concerns. These concerns relate to the observability of subjective class identity when using traditional items of class measurement [
29] (p. 610). One’s racial identity is a variable that matters in subjective class identification [
27].
The full title of Williams’ [
3] book is White Working-class: Overcoming Class Cluelessness in America, but class and race are not explored in a way that elucidates their complex relationship in the American context. We hear plenty of references to “working-class”, but discussion of white workers is fleeting. Non-college educated Whites made up 63% of the Trump coalition [
6] (p. 14). Consequently, we must look elsewhere if we are to clarify the role of race and class in America. Greater clarity in this respect comes from Reed [
30]. Reed notes that the juxtaposition of class and race so ‘familiar… in debates about American inequality’ misunderstands both phenomena by ‘treating them as… indistinguishable’ [
30] (p. 266). Discussing class in a vacuum errs on the side of ‘simplistic, economistic interpretation’ [
30] (p. 270). However, this is problematic, for it disregards the importance of the role race plays in class struggles in the United States. Indeed, such thinking was endemic of the ‘inability’ of large parts of the American left to think of race and class together during the 2016 campaign season [
31] (p. 2). Bernie Sanders, for instance, who ran to the left of Hillary Clinton in the Democratic primaries, consistently emphasised that class was more important than race and identity politics [
32]. Clinton, who was more embracive of the role that ethnic identity plays in class struggles, resultantly won the nomination.
A way of thinking about race and class that brings the two into one frame comes from Roediger [
33]. In Wages of Whiteness, Roediger [
33] argues that antiracist identity politics are a just response to the “racialisation” of class politics. Wages of Whiteness thus sets the foundations for critical whiteness studies to note how the category of “working-class” became intertwined with connotations of race. As Virdee [
31] argues, to see oneself as working-class was also to see oneself as white and in ‘relational opposition to… non-White social groups’ (p. 2). To authors such as Roediger [
33], race is thus not a false construct of ideas and beliefs, but a simulacrum with a basis in reality. Consequently, we now see how race and class are better understood when construed as ‘equivalent and overlapping elements’ rooted in a ‘singular system of social power and stratification’ [
30] (p. 266).
Race and class have a ‘historically specific’ meaning in America [
30] (p. 272). Their intersection is a ‘fact of life’ older than the Republic itself [
30] (p. 266). Examination of this history is absent in the works of class in America we have discussed thus far. Consequently, we now turn to an authoritative past voice on the subject. An authoritative text on the history of whiteness and labour is Black Reconstruction by Du Bois [
34]. The “White Worker” that Du Bois posits reaps the monetary benefits ascribed by their class status. While a position predicated on racial disparities had prevailed since the early time of the Republic, the institution of involuntary servitude had started to weaken by the mid-Nineteenth Century.
In 1857, anti-slavery fervour was catching in the English labour movement. However, such initiative had ‘limited influence’ across the Atlantic [
34] (p. 25). American unions were willed to abolish servitude, but ‘[presently] self-preservation called for slavery’ [
34] (p. 25). In other words, unions expressed concern at the prospect of millions of poor White labourers competing for jobs with free slaves. Indeed, poor Whites expressed the ‘vivid fear of the Negro as a competitor in labor [sic]’ [
34] (p. 29). While Wilson [
35,
36] questions the relevance of race around the economic arrangements of contemporary American society, the election of Trump prompts a re-evaluation. Academics and commentators debating whether economic insecurity or racial animus drove Trump’s election have missed whether, in a Du Boisean vain, it is some alchemy between the two [
37].
Now that we have considered three factors in the determinants of working-class whiteness—education, income and racial self-identification—we begin to see the multidimensional nature of the sociodemographic group. By extension, we also begin to see the complexities present in the formulation class identities at both the individual psychological level and for the measurement of the group in empirical research. An understanding of these multiple facets is especially vital regarding the later; to arrive at a definition of the White working-class that will serve as a basis for the subsequent measurement of the sociodemographic group, we now see that it is formulated by drawing from a combination of these factors. Notwithstanding, we are at a loss as to the relative weight of these factors in the formulation of a White working-class variable. Consequently, the next subsection provides a multinomial logistic regression analysis to see which factors the determinates of class identity for White Americans are.
2.3. Modelling the Predictors of Working-Class Whiteness
The last section delineated the three theoretical underpinnings of white Americans’ identification with working-class strata. These are income, educational attainment, and a more subjective elucidation of class based on one’s racial identity. Based on this theoretical review of the White working-class identity literature, The Author hypothesises that either income, education, white racial identity (or some probable combination of the three) predict working-class whiteness. To test this hypothesis, the Author ran a multinomial logistic regression. Multinomial regression analysis was chosen to model the predictors of working-class whiteness. This method was chosen over binomial logistic regression because the dependent variable—self-identified class status—has four reference categories in the ANES dataset; these are lower class, working-class, middle class and upper-class.
2.3.1. Procedure
A sample population was drawn from the American National Selection Study (ANES) 2016 Time Series dataset. The 2016 Time Series Study was chosen because the dataset has a sufficiently large sample size (n = 4271) to conduct the analysis. Additionally because ANES asks questions relating to White racial self-identification that other surveys such as the General Social Survey (GSS) does not. Without the addition of a White racial identification variable in the model, we would not be able to see if the construct is a predictor of working-class whiteness. Cases from the 2016 Time Series Study were selected if they met the criteria for being White and native-born. For race, cases were included if they were coded as non-Hispanic White. Meanwhile, native-born Americans coded as either being born in one of the 50 U.S. states or Washington D.C. were selected for inclusion.
After the population had been selected, the dependent and independent variables were then chosen. For the dependent variable, a question asking respondents to self-identity with one of four classes was selected. Next, three independent variables were selected for income, education and racial self-identification. For household income, respondents were given a range of incomes and asked which bracket their annual pre-tax income fell. For education, respondents were asked their highest level of educational attainment. For racial self-identification, respondents were asked on a 5-point Likert scale how important being White was to their identity. Certain variables with large amounts of categories, such as income and education, were also recoded to simplify the number of categories. The income variable was reduced from 26 categories to seven categories. The educational attainment variable was reduced from 16 categories to six. The descriptions and ranges for all of these variables can be found in
Table A1. In case of the need for control variables in the analysis, three supplementary sociodemographic variables that included sex, age, marital status as well as one socioeconomic variable, home ownership status, were selected for inclusion. Cases were excluded for nonresponse to and refusal to answer to ensure a complete dataset. After case selection and data cleaning, there were
n = 1335 valid cases from an initial total sample of
n = 4271.
2.3.2. Assumption Testing
Before the multinomial logistic regression was conducted, assumptions had to be met regarding the suitability for the data for the analysis. Tests for multicollinearity were conducted to check that the independent variables were not highly correlated with one another. The Pearson Correlation Coefficient Test indicated that the independent variables were weakly correlated with one another, with the highest correlation being between income and educational attainment at 0.371. The full correlation matrix can be found in
Table A2 in
Appendix A. A further inspection of the Tolerance and Variance Inflation Factors (VIF) also indicated no multicollinearity present between the independent variables. The lowest score for Tolerance was income at 0.855 (a Tolerance value of less than 0.100 would indicate multicollinearity). Likewise, the highest VIF was also for income at 1.169 (a VIF of 10 or more would indicate multicollinearity). The full collinearity statistics can be found in
Table A3 in
Appendix A. After checking for no multicollinearity, tests were conducted to detect any outliers in the data. Cooks Distance and a test of Leverage Distance indicated no significant outliers within the dataset.
2.3.3. Model Fit
Three tests were conducted to assess overall goodness-of-fit and how well the model fitted the data. The first two tests conducted were Pearson’s Goodness-of-Fit Test and the Deviance Test. As the Pearson and Deviance tests are a measure of how poorly the model fits the data, we would expect the tests to not be statistically significant to indicate a good model fit. Both of the tests indicated that the model was a good fit for the data. For the Pearson Test, X2 (495) = 411.828, p = 0.997. For the Deviance Test, X2 (949) = 377.429, p = 1.000. For the third test, model fit was assessed using the likelihood-ratio test. The likelihood ratio test works by analysing changes in model fit when comparing the full model to the intercept-only model. The difference between the −2 Log Likelihood of the intercept-only model and the full model has a X2 distributed with degrees of freedom equal to the difference in the number of parameters. The model fit for the intercept-only model had a −2 Log Likelihood of 1450.141, while the full model had a −2 Log Likelihood of 742.640. The greater the difference between the two values, the better that income, education and White racial self-identification are in explaining class identity. This difference was 707.501, which is Chi-Square distributed with 45 degrees of freedom and is statistically significant p = 0.000. The final model thus statistically significantly predicted the dependent variable over and above the intercept model, X2 (45) = 707.501, p = 0.000. Overall, the model successfully predicted 62% of cases.
2.3.4. Results
Once all of the assumptions regarding the suitability of the data had been met, a multinomial logistic regression was performed to model the relationship between the independent predictor variables and native-born White American’s self-identification with working-class strata. First, a Likelihood Ratio Test was performed on the model to test which of the three predictor variables were statistically significant (See
Table A4). The test showed that income and educational attainment had a statistically significant effect (
p = 0.000). However, White racial self-identification reported a
p-value of 0.282; meaning that the variable had no discernible statistically-significant effect on predicting working-class self-identification. Next, the accepted 0.005 benchmark for statistical significance was employed for all three predictor categories (See
Table A5) As evidenced by
Table A5, the first four predictor categories of household income all had a statistically significant effect on predicting self-identification with working-class strata. The last two income categories (total household income of
$100,000 or more) had no statistically significant effect. Across all education categories except the graduate category, there was also a statistically significant effect on predicting the dependent variable (
p = 0.000). Consistent with the findings of the Likelihood Ratio Test, none of the categories within the white self-identification variable found a statistically significant effect on the dependent variable, with
p values ranging between 0.074 and 0.674.
2.3.5. Discussion
These results indicate that education and income have a greater predictive power in accounting for working-class membership among native-born White Americans while racial identity does not. The results from the education variable perform as expected; we would not, for example, expect those with a higher level of educational attainment to identify strongly as working-class white, since graduates have higher average lifetime earnings than those with a lower level of attainment. The model shows this, with the odds of identifying with working-class strata decreasing as White Americans’ level of education increases. The income variable is more significant than education in predicting working-class whiteness. White Americans in households making less than
$100,000 annually report a stronger association with working-class group membership. Among those making between
$0–
$74,999, the odds ratios are especially high and likewise increase as household income level decreases, except in the case of the lowest income bracket. (
Table A5) In respect to predicting what makes one identify as working-class white, White Americans in households making less than
$100,000 is more important than levels of educational attainment or how important they perceived their whiteness to be. This finding could prove to be especially significant if it can be determined that income plays a factor in how White Americans wish to be perceived. This is to say that there may be credence to the idea that being perceived as working-class gives more credibility as an individual, even if one’s income is sufficiently high enough to not make them “working-class”. Such a finding would have important sociological implications, but further analysis is beyond the scope of the current paper.
While these findings are somewhat significant, we must exercise caution because the model only accounted for 47% (Nagelkerke 0.470) of the total variance of the dependent variable. This means that there are other predictors of working-class whiteness not accounted for in the analysis. These findings provide us with a more sophisticated and nuanced understanding of the determinants of class status among working-class whites. However, there is still further analysis that can be done if we are to fully understand “working-class white” as a variable. Given that we have party accounted for working-class whiteness by means of a regression analysis, we can now contextualise the findings of the literature within this new definition.