These data are derived from the Census Bureau’s Current Population Survey Voting and Registration Supplement (CPS).
Please see the methodology statement at this bottom of the page for more information on how these statistics are calculated.
Race and Hispanic Ethnicity
Turnout rates for four racial and Hispanic ethnic groups are graphed here: (1) Non-Hispanic White, (2) Non-Hispanic Black, (3) Non-Hispanic Other, and (4) Hispanic. All other racial groups are combined primarily because of relatively small Asian, Native-American, and other groups in the early years of the series.
Non-Hispanic White Share of the Electorate
The United States population is becoming more diverse. The Census Bureau estimates that some time in the 2040’s, Whites will no longer be the majority of the population due primarily to differential death and birth rates. These slow-moving forces are reflected in the declining share of the non-Hispanic White population within the electorate starting in the early 1990s as more diverse younger people become voters. This trend in the electorate will take some years to catch up with the general population because older people are more likely to vote than younger people (see below).
Turnout rates for four age groups are presented here: (1) 18-29, (2) 30-44, (3) 45-59, and (4) 60 or older. These age ranges are chosen for presentation purposes.
Turnout rates for four education categories are presented here: (1) not a high school graduate, (2) high school graduate, (3) attended some college or graduated from college, and (4) post-graduate studies.
Most states do not provide demographics from their voter registration files, so we must turn to surveys to understand the demographic composition of the electorate.
The venerable Census Bureau’s Current Population Survey Voting and Registration Supplement is a frequently-used survey because of the large sample size and long time series dating back to the 1960s. These statistics presented here, however, are different from those published by the Census Bureau, and are likely more accurate.
The Current Population Survey is a survey, and like others it has errors associated with the familiar margin of error as well as other less well-known errors, such as non-response bias and vote over-report bias.
Survey respondents rarely answer all questions, and this is particularly true for the the CPS Voting and Registration Supplement, which follows a lengthy questionnaire on household demographics and employment (the primary purpose of the monthly CPS survey is calculating the nation’s unemployment rate). It is standard practice for survey researchers to count these non-respondents as missing data, but the Census Bureau counts them as if they responded they “did not vote.” An effect of this practice is that the Census Bureau can estimate a number of voters to be close to the election results. This is not standard practice and can lead to misperceptions of turnout. For example, Chief Justice Roberts wrote in his Shelby County decision how racial discrimination in voting was in the past because African-American turnout in Mississippi exceeded Massachusetts. However, this was an artifact of fewer Massachusetts African-Americans responding to the Voting and Registration Supplement.
If the Census Bureau followed standard practice of coding “did not vote” as missing data, then the CPS turnout rate would be increased by ten or more percentage points. This is because of another phenomenon known as vote over-report bias, where for whatever reason, more people say they voted than official government reports indicate. This is a widely-observed phenomenon across all election polls.
To account for non-response and vote over-report bias, I reweight the CPS using a procedure proposed by Hur and Achen (2013). This procedure effectively reweights the CPS state level turnout rates to be equal the voting-eligible population (VEP) turnout rates published here.
This isn’t a perfect solution, but it is probably yields turnout rates closer to the truth. Although a minor issue, the Current Population Survey’s universe is the non-institutional citizen population of the United States, which is a relatively close approximation to VEP. A primary difference is that “non-institutional” populations include dorms, military barracks, nursing care, and prisons. For most states only prison populations are not included in the VEP.
For more information see data and documentation.