Absolutely Relative: How Education Shapes Voter Turnout in the United States

Campbell (2009) did find evidence of the effect of relative education on other types of engagement, including persuading others; displaying buttons, signs, or stickers; contributing financially to campaigns; and volunteering for candidates or political organizations.

This is the average total tuition, fees, room, and board charged for full-time undergraduate students in degree-granting institutions. The values are based on the Consumer Price Index (CPI), adjusted to the academic year 2017–2018. From 1985–1986 to 2017–2018, the average price of an education at a public university increased from $8,143 to $17,797.

Scholars often point out that CPS voter turnout may be overreported (Ansolabehere and Hersh 2012; Ansolabehere et al., 2022). The CPS relies on self-reports of voting, and it has been suggested that certain respondents, including blacks and Hispanics, may overstate turnout rates (Ansolabehere et al., 2022). While this is a limitation of the data available, there are at least two reasons why this study uses the CPS data. First, this study assumes that the overreporting issue does not threaten this study’s inferences, because there is no evidence that the overreporting issue caused by individuals’ educational attainment has increased over time (Ansolabehere & Hersh, 2012). Ansolabehere and Hersh (2012) examined reported voter turnout in the 1980, 1984, and 1988 National Election Study (NES) and voter turnout in the 2008 Cooperative Congressional Election Study (CCES) and found that the correlation between individuals’ educational attainment and overreporting of their voter turnout has decreased over time. In fact, there was no evidence suggesting inflated overreporting of voter turnout caused by individuals’ educational attainment over time (Ansolabehere & Hersh, 2012). Second, the CPS covers the longest timeframe available, capturing individual voter turnout in the United States from 1976 to 2018.

Previous studies on the relative education model limit their analyses to respondents age 25 or older (Campbell, 2009; Nie et al., 1996). This study also assumes that individuals are likely to have finished their education by the time they reach age 25, meaning that by this point the full effect of their absolute and relative education may be analyzed.

This research only considers eligible voters. The variable is coded 1 if respondents voted in the most recent November election or 0 otherwise.

The original educational attainment variable from the CPS is a combination of two other variables. The first variable is the respondent’s highest grade of school or year of college completed, with data available until 1992. Afterward, the CPS introduced a new measure of educational attainment that captures the highest degree or diploma attained by the respondent. To reconcile the two variables, this study follows Jaeger’s (1997, 2002) method of imputing the highest grade completed. For example, this study imputes 2.5 years of education for those who answered that their highest grade or degree completed is “1st, 2nd, 3rd, or 4th grade” and 5.5 years for respondents whose highest grade or degree completed is “5th or 6th grade.” Please see Jaeger’s (2002) Table 2 (p. 10) for more detailed information on how to change the categorical responses to the highest grade completed.

The state is identified as the household’s state of residence in the CPS data. Campbell (2009) utilized three levels of geographical data: the state, the metropolitan area, and the zip code. Unfortunately, the metropolitan area data has only been available since the 1990s, and the zip code information is not available in the CPS data. Although this paper attempted to measure relative education using the county as the geographical unit based on Federal Information Processing Standards (FIPS) county codes, these data are only available beginning in 1996. Given this condition, this study uses the best available geographical unit, the state, to capture individuals’ social networks.

Table 2 in Appendix 1 presents descriptive statistics for all the variables used in this paper. By including age and age squared, this paper separates the relative effect of education from the effect of age to overcome the limitations of the relative education measure that Tenn (2005) and Campbell (2009) criticized. The gender variable is a dichotomous variable, coded 1 for female and 0 otherwise. The race variable is coded as a categorical variable with whites as the reference category. Marital status is coded as 1 for married respondents and 0 otherwise. As the CPS lacks data on partisanship, which has often been included in the voter turnout literature, this study could not control for the strength of respondents’ partisanship. Lastly, this study does not include respondents’ personal income for two reasons: (a) many studies on the relative education model have not controlled for income (Helliwell & Putnam, 2007; Nie et al., 1996; Persson, 2012; Tenn, 2005) and (b) there are no data on personal income in the CPS Voting and Registration Supplement, and other financial indicators, such as total household income or family income, are either unavailable or available only for limited time periods.

Estimates here are produced using the fixest function in the R package, Fast Fixed-Effects Estimations (see more detailed information https://​cran.​r-project.​org/​web/​packages/​fixest/​fixest.​pdf).

In Appendix 1, Table 3 shows the results of a bivariate analysis on the relationship between the absolute value of education and voter turnout (Model 5) and the relationship between the relative value of education and voter turnout (Model 6). The results show that an additional year of education increases individuals’ propensity to vote, and individuals with higher educational positions are more likely to vote.

Marginal effects and standard errors of the interaction models were calculated following Aiken and West (1991).

Midterm elections show consistently lower voter turnout than presidential elections in the United States; therefore, previous studies have emphasized the importance of analyzing voter turnout separately for each type of election (Beer, 2017; Tolbert & Smith, 2005).