How Many People Voted? Turnout Depends on the Denominator

With midterm elections on 3 November, headlines will soon announce a turnout rate, and one question will go unasked: turnout out of whom? The same number of ballots can be described as a low, middling or high percentage, depending on a choice most reports never mention. It is a lesson in how denominators shape statistics, and it has nothing to do with which side you support.

Three denominators

The voting-age population (VAP) counts everyone 18 or older, which includes people who cannot vote, such as noncitizens and some people with felony convictions, and excludes eligible citizens living overseas. The voting-eligible population (VEP) adjusts for those groups. Registered voters is the smallest group, because it includes only those who have signed up (UCSB American Presidency Project).

One election, three answers

Imagine a state with a VAP of 7.6 million. Subtract 0.6 million noncitizens and 0.2 million other ineligible adults, add 0.1 million eligible overseas citizens, and the VEP is 6.9 million. Suppose 5.5 million are registered and 3.9 million cast ballots. Turnout is 51.3% of the VAP, 56.5% of the VEP and 70.9% of registered voters. Same election, three headlines. The scientific calculator does the division.

This is more than an academic point. McDonald and Popkin argued that the apparent decline in turnout since 1972 was largely an illusion created by using voting-age population as the denominator, since the ineligible population, not nonvoting, had been growing (McDonald and Popkin, 2001).

Choose the denominator before you read the percentage.

Surveys have their own problem

Turnout is also measured by asking people. Suppose a poll of 1,000 adults finds that 68% say they voted. The confidence interval for a proportion runs from 65.1% to 70.9%, yet the official count implies 56.5% among eligible adults. The one-sample proportion test returns a z-statistic near 7.3.

A gap that large is beyond sampling error, which points to bias. Researchers have documented over-reporting of turnout in survey data (McDonald, Political Analysis). People say they voted partly because it feels like the right answer.

Precision needs sample size

Even an honest survey needs enough people. To estimate turnout near 56.5% within ±3 points requires roughly 1,050 respondents, and ±1 point requires about 9,400. The sample size calculator gives these numbers, and our article on poll margins of error explains the arithmetic.

A checklist for headlines

Ask which population is in the denominator, whether the numerator counts all ballots or only some, whether the figure is official or survey-based, and which definition the comparison year used. A "record low" under one definition can be a "record high" under another. Always print the denominator beside the percentage, as in "56.5% of eligible voters."

The numerator moves too

Early headlines rely on incomplete counts, and totals typically keep rising as late ballots are counted and results are certified. A turnout figure published on election night is therefore a preliminary estimate. Compare final numbers with final numbers, and midterm elections with midterm elections, since presidential years draw different electorates.

The same care applies across states and countries. Systems with automatic registration, compulsory voting or different eligibility rules define denominators differently. None of this says whether participation is healthy. It says only that the number needs a label.