Relative Education or Compensation?

For decades, researchers have pointed to educational attainment as a strong correlate of political participation (R. E. Wolfinger and Rosenstone Reference Wolfinger and Rosenstone1980). Despite rising average education levels among younger generations, education-driven voting patterns observed in the mid-twentieth century bear a striking resemblance to contemporary voting patterns (Leighley and Nagler Reference Leighley and Nagler2013). A resulting question is whether educational attainment causes voting behavior. Some research supports the idea of a causal link (e.g., Dee Reference Dee2004; Sondheimer and Green Reference Sondheimer and Green2010), while some question it (e.g., Berinsky and Lenz Reference Berinsky and Lenz2011; Kam and Palmer Reference Kam and Palmer2008).

In their relative education model, Nie, Junn and Stehlik-Barry (Reference Nie, Junn and Stehlik-Barry1996) advance a view of education as a competitive status indicator rather than a socializing agent causing participation. In their view, education reflects a set of social advantages, chief among them a position in the center of well-connected social networks. Individuals with high levels of educational attainment, compared to their age-group peers, participate in politics both as a reflection of their elevated social status and as an effort to preserve that status by working to influence political decisions.

Their model implies that one’s education relative to their local social and political networks—also referred to as their “educational environment”—should predict participation better than one’s absolute level of educational attainment. As Campbell (Reference Campbell2009) succinctly puts it, “having a college degree in a place where few people have college degrees provides a larger boost to engagement than being college-educated in a place where college degrees are the norm” (p. 777). Indeed, several studies suggest higher relative education is associated with greater political participation over and above any association with educational attainment (Campbell Reference Campbell2009; Persson Reference Persson2011; Tenn Reference Tenn2005).

The relative education model provides important insights that education reflects a competitive social advantage and that the probability of voting does not increase monotonically with educational attainment. However, it does not account for the heterogeneous effects of education, and civic education in particular, on participation across subgroups. Civic education appears to more strongly affect political socialization among young people from disadvantaged groups, exerting what Campbell (Reference Campbell2019) calls a “compensation effect” (see also Campbell Reference Campbell2008; Langton and Jennings Reference Langton and Jennings1968). Students from lower socioeconomic backgrounds demonstrate greater political engagement if exposed to more civic education in school (Lindgren, Oskarsson, and Persson Reference Lindgren, Oskarsson and Persson2019; Neundorf, Niemi, and Smets Reference Neundorf, Niemi and Smets2016). Mandatory high school civics examinations increase political knowledge among adolescents from Latino and immigrant families (Campbell and Niemi Reference Campbell and Niemi2016). Some techniques appear to be particularly effective. While talking openly about politics in the classroom is associated with greater engagement broadly (Martens and Gainous Reference Martens and Gainous2013; Torney-Purta Reference Torney-Purta2002), specific conversations that highlight the political agency of marginal groups or validate their historic distrust in government do more to engage young people from marginal backgrounds (Nelsen Reference Nelsen2021, Reference Nelsen2023).

Of course, educational attainment includes more than civics education alone. However, it is plausible that students with higher levels of educational attainment have also been exposed to more civic education. Moreover, some authors argue that aspects of education outside of civic education, such as noncognitive skill development and the internalization of social norms, explain the connection between education and participation (Hansen and Tyner Reference Hansen and Tyner2021; Holbein and Sunshine Hillygus Reference Holbein and Sunshine Hillygus2020). Though the content of civic education may help to explain group heterogeneity, it is enough for our purposes to say that something about the educational system in America produces differing relationships between education and participation across groups.

The relative education model should also reflect group differences. If education is more strongly related to participation among disadvantaged groups as a result of its socializing effects, then we should expect to see that relationship independent of a person’s educational environment. In other words, the members of disadvantaged groups should participate more as a result of their education, regardless of their educational environment.

Translating this argument to racial and ethnic groups, we contend that the relative education model will better capture the relationship between education and voting among White citizens than among Black or Latino citizens. For a number of historical reasons, including but not limited to slavery, economic exclusion, and the inequitable provision of public goods such as education, generations of African Americans and Latinos had limited opportunities to participate in American politics. Political inequality is reproduced across generations (Brady et al., Reference Brady, Lehman Schlozman and Verba2015). The transmission of knowledge, values, and norms surrounding civic engagement often occurs in families and communities (Gimpel et al., Reference Gimpel, Celeste and Schuknecht2003; Verba, Burns, and Schlozman Reference Verba, Burns and Schlozman2003). Children whose parents had fewer opportunities and resources to become politically engaged tend to be less engaged themselves. Therefore, as a result of historical exclusion, we expect Black and Latino youth to have fewer opportunities to become socialized into voting outside schools, with schools serving as important forays into civic engagement. For Whites, who have more ample opportunities for political socialization outside schools, we may reasonably expect educational attainment to better reflect relative social status.

Then again, many opportunities for political socialization for Black and Latino youth are available outside schools. Aside from families, campaigns and community organizations like churches have also helped to mobilize Black and Latino voters (Barreto, Segura, and Woods Reference Barreto, Segura and Woods2004; Walton Reference Walton2020). Moreover, many civically active youth see nonvoting forms of participation, like protest, as more efficacious, weakening the connection between socialization and voting (Searles and Suzuki Reference Searles and Suzuki2024). While we expect to observe evidence more consistent with the compensation effects of education among these groups, it is possible that these alternative pathways to participation mitigate any such effect.

We formalize our expectations below:

For Asian American voters, we hold weaker expectations given the wide variety of paths to political socialization and more splintered efforts at mobilization (Wong et al. Reference Wong, Karthick Ramakrishnan, Lee, Junn and Wong2011). On one hand, relative socioeconomic advantages among this group would predict a relationship between educational attainment and voting reflective of the relative education model. On the other hand, Asian American voters tend to vote at lower rates despite their socioeconomic advantages and benefit less from civic education interventions than their Black and Latino peers (Fraga Reference Fraga2018; Nelsen Reference Nelsen2023). Therefore, we explore the moderating effect of educational rank for this group without formulating a specific expectation.

To the best of our knowledge, Campbell (Reference Campbell2009) provides the only prior test of racial differences in the relative education model and finds no difference in its ability to describe nonvoting forms of participation among White and Black Americans. However, Campbell’s results are derived from an analysis of a cross-sectional survey taken more than 20 years ago, and the results do not describe voting behavior. Leveraging more expansive data over time may yield different results. Using such data, Kim (Reference Kim2023) finds that the absolute model of education better describes the relationship between education and voting than the relative model after 2000. Though Kim (Reference Kim2023) points to widening economic inequality as an explanation, we note that the years after 2000 have also been marked by greater racial diversity in the electorate.

Since Nie et al. (Reference Nie, Junn and Stehlik-Barry1996) articulated their model nearly 30 years ago, it is important to test it in an increasingly diverse electorate. As research has shown that political socialization operates differently for minoritized populations compared to Whites (Raychaudhuri Reference Raychaudhuri2020), so too may education. We acknowledge that education might have heterogeneous effects on participation within groups as well. Rather than delving into the many mechanisms through which educational attainment may impact participation, we are interested in mapping how this relationship varies systematically across racial groups despite within-group variation. Understanding the differences between groups can help us better understand the role that education may play in addressing turnout gaps.

Data and Methods

This analysis uses the CPS Voting and Registration Supplement, which includes self-reports of election participation for presidential and midterm elections over several decades. We use CPS data from 1978 to 2020, during which educational attainment was measured in all states. The sample includes over 1.8 million respondents, with approximately 76,000 to 97,000 respondents per survey year. The large sample gives us a sufficient number of responses to analyze subgroups and measure respondents’ educational attainment relative to birth and geographic cohorts.

We follow prior work and model turnout as a function of the interaction between an individual’s own education and their educational rank within a defined age-place cohort (Campbell Reference Campbell2009; Kim Reference Kim2023). For comparison, we estimate both an additive model (1) and a relative education model (2), defined below. The additive model serves as a test of H1 and H2, while the relative education model serves as a test of H3a-c.

(1) $$\eqalign{ & {\rm{Voter\ Turnou}}{{\rm{t}}_{it}} = {\beta _0} + {\beta _1}{\rm{ Educational\ Attainmen}}{{\rm{t}}_{it}} + {\beta _2}{\rm{Educational\ Ran}}{{\rm{k}}_{it}}{\rm{ }} + \cr & {\beta _{3}}{\rm{Controls}}{\rm{ }} + {\epsilon _t} \cr} $$

(2) $$\eqalign{ & {\rm{Voter\;Turnou}}{{\rm{t}}_{it}} = {\beta _0} + {\beta _1}{\rm{Educational\ Attainmen}}{{\rm{t}}_{it}} + {\beta _2}{\rm{Educational\ Ran}}{{\rm{k}}_{it}} + \cr & {\beta _3}{\rm{Educational\ Attainmen}}{{\rm{t}}_{it}}{\rm{X\ Educational\ Ran}}{{\rm{k}}_{it}} + {\beta _4}_{\rm{Controls}} + {\epsilon _t} \cr} $$

The dependent variable (Turnout) is dichotomous, indicating whether the respondent i reported voting in the most recent election of survey year t. Self-reported voting measures raise perennial concerns due to challenges with overreporting (e.g., Ansolabehere and Hersh Reference Ansolabehere and Hersh2012). Recent findings that CPS data produces biased estimates of turnout, particularly for Black and Latino respondents, compound these concerns for our study (Ansolabehere, Fraga, and Schaffner Reference Ansolabehere, Fraga and Schaffner2022). In the absence of more comprehensive data or CPS validation studies allowing a correction method to be developed, we proceed by analyzing these data but caution readers about known biases in them.

The main independent variables are educational attainment and educational rank. Educational Attainment is a count of years of schooling that the respondent had completed at the time of the survey, following Jaeger’s (Reference Jaeger1997) recommended coding scheme for CPS data. It ranges from 0 (no education at all) to 18 years (Ph.D. or professional degree). Figure 1 illustrates that educational attainment has varied widely across states and over time.

Figure 1.

Educational Attainment Across States (1978 vs 2020).

We measure Educational Rank as a respondent’s percentile of educational attainment among other respondents within their generational cohortFootnote ¹ in their state.Footnote ² The variable ranges from 0 to 1 and captures how respondents’ absolute value of education compares to those in their educational environments (see Helliwell and Putnam Reference Helliwell and Putnam2007; Tenn Reference Tenn2005). By defining educational environments within age cohorts, we eliminate the threat of multicollinearity by including age and educational rank in the same model, a problem in earlier research (Kim Reference Kim2023). For example, a Millennial in 2020 with an undergraduate degree (16 years of education) ranks at about the 41^st percentile of education compared to other Millennials living in Washington, D.C., but ranks in the 83^rd percentile compared to other Millennials in Mississippi. On the other hand, a Baby Boomer with a high school diploma (12 years) in 2020 ranks at the 34^th percentile compared to other Baby Boomers in West Virginia but ranks at the 18^th percentile in Washington state. Thus, respondents’ educational rank varies substantially by generation and geography.

Our model also controls for age, age squared, gender, and marital status, consistent with other scholarship examining relative education (Kim Reference Kim2023; Nie, Junn, and Stehlik-Barry Reference Nie, Junn and Stehlik-Barry1996). In particular, age is positively associated with turnout (R. E. Wolfinger and Rosenstone Reference Wolfinger and Rosenstone1980), women are more likely to vote than men (Leighley and Nagler Reference Leighley and Nagler2013), and married adults are more likely to turnout (N. H. Wolfinger and Wolfinger Reference Wolfinger and Wolfinger2008). We also use fixed effects for years to control for election-related trends. We analyze respondents aged 25 and older since most people have completed their education by that age.

Finally, we code CPS data to place respondents in one of the following groups: White, Black, Latino, and Asian. The White, Black, and Asian categories include those who solely identified with that racial group. Latinos are those who indicate that their ethnicity is Hispanic or Latino, regardless of race. Our sample in total includes 82.8% identified as White, 9.5% as Black, 5.5% as Latino, and 2.2% as Asian.Footnote ³

Findings

We begin by illustrating our two independent variables of interest: Educational Attainment (Figure 2a) and Educational Rank (Figure 2b) plotted against average self-reported turnout scores. Rank values are binned at every fifth percentile. As the two figures illustrate, education in both absolute and relative terms is strongly and positively correlated with voter turnout.

Figure 2.

Average Turnout by Educational Attainment and Race.

As a test of H1 and H2 using the additive model, we estimate logistic regression models with robust standard errors for each group, with and without controls. Table 1 presents the results. As expected, both educational attainment and educational rank are significantly and positively related to turnout across all groups in the fully specified models. Therefore, we find support for H1 and H2.

Table 1.

Effects of Educational Attainment and Rank on Voter Turnout by Race/Ethnicity

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

Next, we turn to our primary analysis by estimating the relative education model with separate models by race/ethnicity to test H3a-c. We estimate a logistic regression model with robust standard errors for each group and include an interaction term between educational rank and educational attainment. We focus our interpretation on the fully specified models in the last four columns in Table 2.

Table 2.

Effects of Relative Education on Voter Turnout by Race/Ethnicity

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

The coefficient estimates for the interaction terms in those models suggest group differences, but they are difficult to interpret on their own. To facilitate interpretation, we plot the estimated marginal effects of educational attainment with 95% confidence intervals over the range of educational rank in Figure 3. In line with expectations, the figure depicting White respondents shows a negative slope, indicating that the marginal effect of an additional year of education on turnout diminishes as one climbs the educational rank ladder. Also in line with expectations, the slopes for Black and Latino voters are flat, indicating their educational environment does not condition the association between educational attainment and voting. Among Asian voters, the marginal effect of educational attainment decreases over the range of educational rank.

Figure 3.

Marginal Effect of Educational Attainment Across Values of Educational Rank.

To present these findings in more substantive terms, Figure 4 illustrates how the predicted probability of voting (y-axis) for each racial group varies as years of educational attainment increase (x-axis) while holding respondents’ rank within their educational environment constant at two different values: the 40^th percentile and the 90^th percentile. For White respondents at the 40^th percentile, moving from a high school diploma to a graduate degree (12 to 18 on the x-axis) results in a roughly 16 percentage point increase in the likelihood of voting. For White respondents at the 90^th percentile, moving from a high school diploma to a graduate degree results in only a 12 percentage point increase in the likelihood of voting. This finding is consistent with expectations; the association between educational attainment and voting for Whites is moderated by educational rank. However, the marginal effect is rather small—the difference in gains from a six-year increase in educational attainment at the two educational ranks is about four percentage points.

Figure 4.

Probability of Voting by Educational Rank and AttainmentNote: Low educational rank = 40^th percentile, high educational rank = 90^th percentile.

For Black and Latino respondents, however, the probability of voting at different levels of educational attainment moves in parallel across the range of educational rank. Among African Americans, the difference in the probability of voting at the 40^th percentile and 90^th percentile of educational rank is roughly 12 percentage points at all levels of educational attainment. Among Latinos, that difference is about 17 percentage points. We see a result similar to Whites among Asian Americans. For those at the 40^th percentile, moving from a high school diploma to a graduate degree results in a roughly five percentage point increase in the likelihood of voting. At the 90^th percentile, the same change in educational attainment yields a roughly two percentage point increase in the likelihood of voting.

As a robustness check, we conduct likelihood ratio tests to compare the fully specified models with and without the interaction between years of education and educational rank. The results, provided in Table A1 in the appendix, indicate that the interaction significantly improves model fit for White and Asian respondents (p < 0.05), but not for Black or Latino respondents.

Therefore, the evidence supports H3a-c that relative education more strongly attenuates the relationship between educational attainment and voting for Whites and more weakly attenuates it for Black and Latino respondents. While we did not formalize a specific hypothesis for Asian Americans, the analysis suggests that the relative education model provides some explanatory power for their voting behavior. In Section B of the appendix, we also present results with an alternative measure of Educational Rank, in which respondents’ rank is measured relative to their state-age cohort within their own group. Using this alternative measure leads to similar conclusions.