Zachary Hertz

Zachary Hertz

Graduate student in political science.

Estimates of Youth Turnout Have Recently Diverged

2 minute read

Published:

With the recent release of the final 2020 Cooperative Election Study (CES) dataset, which includes vote validation, I returned to an earlier project examining youth voter turnout. In looking at voter turnout among adults under the age of 30, I noticed a pronounced gap between estimates from the Center for Information & Research on Civic Learning and Engagement (CIRCLE) and the CES.

CES and CIRCLE estimates of youth voter turnout have been fairly close, but differed by 12 percentage points in 2020

There are three different ways to measure turnout using the vote validation variables in the CES. I calculated CES estimates of voter turnout using the first method, coding the unmatched as non-voters. I collected CIRCLE estimates of youth voter turnout for 2020 and 2016, 2018, 2014, 2012, and 2010 from press releases. I then calculated the absolute value of the difference between the CES and CIRCLE estimates, then plotted the results.

We see that the 2010 and 2012 youth voter turnout estimates were incredibly close, with a difference of just 0.2 and 0.1 percentage points respectively. These differences widened slightly between 2014 and 2018 but remained within about 3 percentage points: estimates differed by 3.3 percentage points in 2014, 1.0 percentage points in 2016, and 2.6 percentage points in 2018. In 2020, however, while the CES validated vote data estimated that voter turnout among adults ages 18-29 was 38 percent, CIRCLE found that 50 percent of adults between 18 and 29 turned out to vote, marking a difference of 12 percentage points.

Potential causes

The 12-point difference in the CIRCLE and CES estimates of youth voter turnout may be the result of methodological differences. CIRCLE states that their estimates are based on voter file data from 41 states — Alaska, DC, Hawaii, Maryland, Mississippi, New Hampshire, North Dakota, Utah, Wisconsin, and Wyoming lack reliable vote history data by age. These states are not omitted from the CES data, however, so unobserved youth voting behavior in the 10 omitted states could lead to different estimates from the different sampling frame.

The CES data guide also notes vote validation matches are only made when there is a high level of confidence that respondents are assigned to the correct record. Therefore, CES records may lack vote validation due to incomplete or inaccurate information. It is also possible that there may be a systemic issue with incomplete or inaccurate voter file data, particularly among those under the age of 30.

You May Also Enjoy

Educational Polarization in the CES

8 minute read

Published:

Recently, CES Researcher Pia Deshpande wrote an excellent tutorial detailing how to plot trends over time using CES data — I highly encourage anyone who hasn’t yet read the piece to promptly do so! This style of plot pairs particularly well with the cumulative CES Common Content dataset created by Shiro Kuriwaki. Together, these resources can help us to understand an issue that has received increased attention in recent months: educational polarization.

Using the CES to Examine Young Voters

5 minute read

Published:

July 1 marked 50 years since the ratification of the 26th Amendment to the United States Constitution, which legally extended the right to vote to those over the age of 18. Young voters have played a vital – at times, even decisive – role in elections since. But decades after the franchise was extended to 18-, 19-, and 20- year-olds, how does their voting rates compare to American adults overall?

Creating Survey Weights in R Using Census Data

14 minute read

Published:

In survey research, the composition of a sample may differ notably from the population being modeled across important characteristics (e.g. race, age, education, party identification). These sampling errors often reflect systematic bias which can pose a threat to accuracy and the researcher’s ability to make inferences using the data, especially if the error is correlated with the variables of interest.